Automatic IETF Minutes

Markdown Version

Session Date/Time: 15 Apr 2026 13:15

WEBVTT

1 "" (0) 00:00:09.659 --> 00:00:29.659 She's a meta type teacher now. Okay, so it is 09:31. Let's go ahead and get started. I I believe everyone was here yesterday, but just in case people have forgotten, we are operating under the IITF Note Well policy, which is policies around things like intellectual propert.

2 "" (0) 00:00:29.659 --> 00:00:39.660 And your behavior, so please be aware of and familiar with this and if you have any questions, you can talk to myself, to Suresh or to our area director Mike.

3 "" (0) 00:00:39.660 --> 00:00:56.160 So yesterday we seem to make some pretty good progress on some of the very core issues we've discussed. If I look back at our agenda, we've covered.

4 "" (0) 00:00:56.160 --> 00:01:16.160 But not closed, everything up here, down to this line here, let me make this a bit bigger for everyone. And so I think what we'd like to do today is to get through the rest of these, and I I would like to try and do so before.

5 "" (0) 00:01:16.160 --> 00:01:17.730 At lunch if we can.

6 "" (0) 00:01:17.730 --> 00:01:37.080 And then if we can return back to the two things we were discussing before, that training issue and also the, the search area and refine the proposals there and try and make a bit more progress with those. I I think that would make this actually a pretty productive meeting. Does that, that sound good to everyone?

7 "" (0) 00:01:38.580 --> 00:01:58.580 And as a reminder, I think we're scheduled to end tomorrow midday, right? So 01:00 01:00 ish whenever people get too hungry to be productive anymore. So let's get going. We have this cluster of five issues that are loosely grouped together as context, so.

8 "" (0) 00:01:58.580 --> 00:02:18.150 Elaine has your hand up? Elaine, go ahead Ellene. So just to clarify, we didn't make any decisions on the topics we talked about yesterday, right? No, so in the ITF we don't make decisions in the meetings. We always We we come to some sort of understanding or or get a direction.

9 "" (0) 00:02:18.150 --> 00:02:33.930 Together with the meeting, but we have to establish consensus on the mailing list. So what we're hoping to do is to come out of this meeting with strong proposals that we can take the mailing list and confirm the consensus there. That makes sense. But we still have.

10 "" (0) 00:02:33.930 --> 00:02:53.930 Sort of like e.g. on one of the things we still have two options. Yes, that's why we want to reserve a fair amount of time. Even if you don't call it decision, like I think we will need to indicate what we wanna send to the mailing list. Exactly. And that's why we need more time to discuss those. I don't want to just have a beauty contest between two different proposals and the list.

11 "" (0) 00:02:53.930 --> 00:03:13.170 That's not terribly constructive. So let's let's I think because these these aren't all the the grouping here is a little looser, but I think it might be good to go through them very briefly and then try and figure out how to to to address them.

12 "" (0) 00:03:13.170 --> 00:03:33.170 We have 1902 which is introduction clarity regarding relation to existing laws, which was not your issue. Then we have 3.1 conformance, which I think is is a more limited issue. Three two respecting preferences which we have a proposal from.

13 "" (0) 00:03:33.170 --> 00:03:51.000 To millerate that, we have, sorry? We have not me. It is me. Again Lauren. Connected to the network, still blame the Cisco. So he's checking out all the rest of us? Yep.

14 "" (0) 00:03:51.000 --> 00:04:11.000 So we have three two. I'm just gonna ignore that respecting preferences. We have distinguishing access and use which is is probably a little further out from the rest of this discussion as well as box collect data for multiple purposes. So I think probably the two to discuss 1st are.

15 "" (0) 00:04:11.000 --> 00:04:19.620 1902 and one 60 and then the remaining three we can handle more in isolation.

16 "" (0) 00:04:19.620 --> 00:04:35.609 So max, do you want to talk this through your thinking on 1902? Maybe I have two. I think it's probably best to sort of sorry, good morning. I think it's probably best to, to skip some of the preamble and, and.

17 "" (0) 00:04:35.609 --> 00:04:53.639 Maybe scroll down, just the proposed resolution, and this has been on the list for a little while, so I assume some people or most people would have read it, but I want to start just by like making sure everyone's looking at the the same language, which is the proposed edit. So I'll just read verbatim out loud in case that's easier for anyone. So.

18 "" (0) 00:04:53.639 --> 00:05:11.789 The the language I've proposed is the expression of preferences is not intended to overwrite legal preferences or restrictions that maybe embodied in websites terms of use parentheses or conquerable page and paren. And in the event of any conflict between those preferences and websites terms of use, the terms of use shall prevail.

19 "" (0) 00:05:11.789 --> 00:05:31.789 As a matter of law. So that's the proposed language. It's clear from the debate in like the last year as well as some of the comments that were in the chat yesterday, that there's like some confusion around the legal implications that are written into the draft, and like specifically the.

20 "" (0) 00:05:31.789 --> 00:05:39.689 Language around like legal obligations. So I'm gonna try to break it down in, in simple terms from my perspective. So like.

21 "" (0) 00:05:39.689 --> 00:05:59.689 We try to set preferences through the drafts or like define what the preferences and the drafts are and what owners may set, but then we say that they have some legal obligations and I'll I'll read again a little bit verbatim here. So the text in the current vocab draft, number five, like the most recent one.

22 "" (0) 00:05:59.689 --> 00:06:19.729 It says it is intended to be used in jurisdictions where expressing preferences results in legal obligations, then there's a little section that I'll skip over and then without prejudice to applicable laws. But it's completely unclear, right? Like in silent on jurisdictions, legal obligations or laws because it's a complex, like there's a lot going on. So.

23 "" (0) 00:06:19.729 --> 00:06:36.959 I'd like us to consider the fact, 1st of all, like websites today set our terms of use through like well documented practice, right? The the terms used in that like bots are included in that.

24 "" (0) 00:06:36.959 --> 00:06:56.959 Practically think and so for setting preferences also for this documentation, right? Or be an attachment mechanism, like what does it mean? And it's not particularly clear to me and I assume it's not particularly clear to anyone that would document and I'd be kind of shocked if we were taking the position as a working group that preferences.

25 "" (0) 00:06:56.959 --> 00:07:14.039 Would override terms. And yesterday there was a conversation around agreements and it seemed pretty clear from the conversation in the room that agreements would have an impact here and preempt the preferences and.

26 "" (0) 00:07:14.039 --> 00:07:29.849 Terms of use are a literal agreement between websites and visitors to that website. That's how they're formulated. So the way I'd lead off really, the question is like, does anyone here think that terms of use wouldn't?

27 "" (0) 00:07:29.849 --> 00:07:46.739 Preempt preference setting. And if that's the case, like, let's talk about it because I think that that's like a pretty fundamental question to to what we're doing. So yeah. So I don't necessarily think it would preempt it, but I can imagine them being written differently for different audiences.

28 "" (0) 00:07:46.739 --> 00:08:03.689 Right, that terms of use is generally directed towards humans who use the site or may or may not be required to agree to it based on the site mechanics. Whereas the preferences are machine readable.

29 "" (0) 00:08:03.689 --> 00:08:22.709 Use presumably by some sort of automated system. And so I think with different audiences, right? Like they might potentially have kind of different, different terms, right?

30 "" (0) 00:08:25.979 --> 00:08:43.139 Terms of use are flexible, right? And they're at individual publishers discretion, right? Like or site owners discretion if, if we want to use the phrase site owner, and that individual choice allows you to be flexible in that way, right? In a term of use, you can say.

31 "" (0) 00:08:43.139 --> 00:08:59.129 Here are my parameters for human visitation, here's my parameters for bot visitation, like those are things you can say and you can adjust or you can say, here's a general prohibition or a general allowance, right? So like, I I don't want to get into like.

32 "" (0) 00:08:59.129 --> 00:09:15.719 What do you feel about right? Someone's just Mike Joneson. Hi.

33 "" (0) 00:09:15.719 --> 00:09:35.669 Thanks mike. Sorry, basically, that's accountable within the terms of use. Like you're able to do that and that exists. So I I don't see that concern especially because I think it leans into actually what is the strength of terms of use there?

34 "" (0) 00:09:35.669 --> 00:09:51.119 But the way I would interpret what you have written here is that a agent that txt would then have to verify that they're understanding, you know, assuming that they're.

35 "" (0) 00:09:51.119 --> 00:10:07.949 We then have to verify their understanding against the terms of use. I think that's the case, yes. Fundamentally, like, like we said yesterday, right? As as the discussion went.

36 "" (0) 00:10:07.949 --> 00:10:23.999 Agreements preempt what's going on in preference setting, but there's NO, like terms of use are not machine readable, and there's NO proposed attachment for terms of use. And so like an automated system interacting.

37 "" (0) 00:10:23.999 --> 00:10:40.019 With this website, like there's would, it would have to be, it would only be able to kind of run after a human had read the terms of use and approved the the running of it. So I'll.

38 "" (0) 00:10:40.019 --> 00:10:56.129 You you've raised a couple different points there, so I'll try to take them and turn. 1st, like underlying to that there's this like assumption that if it's not machine readable, then it doesn't matter.

39 "" (0) 00:10:56.129 --> 00:11:12.779 Which isn't how, like, unfortunately, like just not how the world works, right? Like, not everything is machine readable, there are other contexts that matter. There's like legal, like jurisdictions again like which we mentioned here in, in the issue is like.

40 "" (0) 00:11:12.779 --> 00:11:28.289 Those exist. There has to be some input into how systems are built. It's not just a matter of like, we set the robots off and they go and do what they want and and that's it, right? Like there is always going to have to be human input regardless of where you do things.

41 "" (0) 00:11:28.289 --> 00:11:46.049 So I'd argue with like that premise to begin with. Second, on machine readability, I don't agree that the terms of use are machine readable, they're more complex to machine read, but we A, have llmiss now that are better at it and second of all, like.

42 "" (0) 00:11:46.049 --> 00:12:06.049 It's relatively simple information that you're looking for, allowing or disallowing. Additionally, in terms of find or to your point on like there's NO pointer, I've proposed in this group before that we could add a pointer that that's relatively easy to do. It would be an.

43 "" (0) 00:12:06.049 --> 00:12:08.339 An optional field, I think.

44 "" (0) 00:12:08.339 --> 00:12:28.339 Aker briefly mentioned it in, in one of his email responses as like a possibility, I don't think it's a difficult thing to do on, but also there are actively pointers to terms of use on every website that I'm aware of, and, I'm not a lawyer, so I'm not gonna cite a s a specific case here and maybe we.

45 "" (0) 00:12:28.339 --> 00:12:38.249 We can come back to it on on list if this is of particular interest to people, but in the EU at least there are court cases that say specifically terms of use being linked at the bottom of a site.

46 "" (0) 00:12:38.249 --> 00:12:56.099 Is enough, like that is enough notice from from a publishers where from like whatever there's also the court case where terms of use was specifically found not to be qualified as machine readable for the purposes of the EU AI opt out.

47 "" (0) 00:12:56.099 --> 00:13:11.369 For the EU AI for the TDM exception, right? Yes. Yeah, I think we need to kind of go through the comments like good point, like let's kind of I think you you're cases on the list for sure. I think Paul, you're next.

48 "" (0) 00:13:11.369 --> 00:13:28.349 Yeah, I think it's similar my concerns to what. I think they are similar to what expressed. So 1st of all, I don't, I don't fully get this like we are here to define a.

49 "" (0) 00:13:28.349 --> 00:13:46.109 Machine readable signals because I think the assumption is that doesn't exist yet. Your underlying claim seems to be, well, we don't have to worry because we have terms of service and we should just look at terms of service.

50 "" (0) 00:13:46.109 --> 00:14:06.109 I I think that runs sort of counter to the purpose of the working group. 2nd thing is, that you asked for sort of examples where machine readable expressions might take preference over terms of.

51 "" (0) 00:14:06.109 --> 00:14:25.769 Service and I think or terms of use, and I think the use situation that was alluded to is at the moment one where that hasn't been fully litigated, but the assumption is that a machine readable expression is the preferred way.

52 "" (0) 00:14:25.769 --> 00:14:41.729 Of, expressing a preference or even expressing a an opt out with legal consequences. The 3rd, and I think I made that on the mailing list is I I also think that some of the drafting.

53 "" (0) 00:14:41.729 --> 00:14:59.429 Is problematic, e.g., the last ***** shall prevail as a matter of law. We don't have a law that we can refer to in this thing. This is a, approach that needs to work in very many different jurisdictions and I don't think.

54 "" (0) 00:14:59.429 --> 00:15:15.809 I would be surprised if anybody in the room here had a full understanding of sort of like the legal ordering between different legal instruments in all jurisdictions that this might apply to. So, a couple, couple things here. So.

55 "" (0) 00:15:15.809 --> 00:15:32.789 Like your your opening salva or statement right is like we are here to make a machine readable thing. Like, I understand that we're here to make a machine readable thing, but we have to be very cognizant of what the limits of a machine readable thing are. And again, like, if agreements.

56 "" (0) 00:15:32.789 --> 00:15:52.789 Are a factor, and we've said yesterday they were and I didn't hear any disagreement to that. So if that's basic disagreement we have I'd like you to like frame it that way. Then let me be clear about that. Like, I don't I think you are sort of pointing to the discussion yesterday where at least I and I assume other people in the room assumed agreements.

57 "" (0) 00:15:52.789 --> 00:16:12.789 Are things that are agreed between parties and I think a lot of people in this room will assume that unilaterally expressed terms of use do not have that quality. So my sort of not non non or even endorsing that thing that agreements override with the preference signals if they are negotiated or.

58 "" (0) 00:16:13.289 --> 00:16:33.289 Agreed between two parties, but like the the expression of terms of use somewhere on a website sort of like does not for me fall into that category of agreements that we were talking about yesterday. So Max, just to like let's collect everybody's thing and we you can respond at the end, I just want to hear everybody out because like otherwise it kind of.

59 "" (0) 00:16:33.289 --> 00:16:47.909 One last thing I forgot, like I also think, let's remember this is in the vocabulary document and this is then very specifically bringing it down to something that happens on websites. So like I think this is again if.

60 "" (0) 00:16:47.909 --> 00:17:07.909 We wanted to address something that needs to be attachment mechanism specific because for a number of other attachment mechanisms, the direct website to websites terms doesn't make sense, right? And I think Max's proposal was based on robotics, so I think your point is well made. And Nate, you're after me on the queue since you put your hand up, ok? So.

61 "" (0) 00:17:08.459 --> 00:17:28.459 Go ahead. I think you're done, right? Yes I'm sorry. So I agree with the others that say that the addition of this to basically say any terms of service override is pretty unworkable. Yesterday I think we were talking about agreements were also specifically in addition to negotiating.

62 "" (0) 00:17:28.459 --> 00:17:47.609 The agreements between parties and not terms of service. I think all of that discussion was about agreements that add to the expressed preferences, not agreements that take away. And so I think generally when you have like a public offering of here's this preference statement, here's what you can do.

63 "" (0) 00:17:47.609 --> 00:18:04.709 You can always add to that agreements, but it's pretty structurally problematic to say we've told you you can do this, but in a different way we take away from that. That's not like a general good drafting strategy when you think about like the way licenses work. The other thing I think is the way terms of service.

64 "" (0) 00:18:04.709 --> 00:18:22.649 That aren't specifically agreed to like affirmatively by the parties are treated is really different. So just in terms of this being understandable and followable, I, I think leaving it silent, like you don't have to say to perhaps override the terms of service, but I think.

65 "" (0) 00:18:22.649 --> 00:18:42.649 In the vocabulary, leaving it silent makes a lot more sense. Landon? Yeah, so actually I was gonna just reiterated, I was gonna say what Paul said at the end, which is a reminder the preferences apply to numerous ways in which content can be retrieved, not just websites, so we don't want to be specific, but I do agree that a state.

66 "" (0) 00:18:42.649 --> 00:18:53.939 Treatment around you know agreements as we talked about yesterday. I think replacing the wording specific to terms of use would be very reasonable because that is what we were focused on.

67 "" (0) 00:18:53.939 --> 00:19:13.939 Thank you. Kevin? Actually I'll pass, I think most of my perspective has been sure. Yeah, I got in line to say something similar to what Meredith right like and and also Paul was saying it's unclear like that like the terms of service will override as a matter of law. So we can make that statement here because we certainly don't know.

68 "" (0) 00:19:13.939 --> 00:19:33.939 Because in some jurisdictions it might or it might not. Maybe the better way is to not say it or like come up with some of the mechanism which says like hey how this is, these two interacts like not defined by this trap might be a better way of saying it because we don't know. I think the 1st part where it says like they're not intending to override the legal preference might be ok, but 2nd part may or may not be. But we can talk about.

69 "" (0) 00:19:33.939 --> 00:19:35.219 That later, so.

70 "" (0) 00:19:35.219 --> 00:19:55.219 But it I don't think we can say it as a matter of fact because we don't know, so Mike? Yeah, I'll take a hard position in part just for clarity of discussion. I think the terms of use and the preferences are totally independent and the.

71 "" (0) 00:19:55.219 --> 00:20:14.339 They apply to different parties. The terms of use apply to human users of the site and in particular those who've created accounts and thereby opted into the terms of use. The preferences apply to crawlers.

72 "" (0) 00:20:14.339 --> 00:20:34.339 And how they treat public content for not logged in interactions. We should not mix the two. Thanks mike. Yeah, I just had a narrow point on this idea that websites or not agreements. We all agreed at the very beginning of this meeting to various rules for the IETF even.

73 "" (0) 00:20:34.339 --> 00:20:41.039 Though we didn't go around and individually sign anything or individually say yes in a response. An agreement can be made.

74 "" (0) 00:20:41.039 --> 00:21:01.039 By one party acting in a way that affirms an offer the other part of the party as a matter of law. And, in fact, we all just did this like 20 min ago. So, not in specific direction of that, but in general in the tone of the discussion I'm just I'm a little concerned that we are wondering in.

75 "" (0) 00:21:01.039 --> 00:21:21.039 Into the area of contract formation in law, which this body is not expert in. I know that terms of service formation on the internet, just in one jurisdiction in the United States is an incredibly complex and dynamic area, so I I everyone should be very careful about, you know, trying to extract that into things that in.

76 "" (0) 00:21:21.039 --> 00:21:25.679 Affect this discussion. So let's try and stay on our.

77 "" (0) 00:21:25.679 --> 00:21:45.089 Topics. Okay, understood. So maybe the idea of a pointer is a good one. Let's just say my preference is you go to this URL. So there is a draft by max. I can send you a pointer to it, you can look at it and comment if you want. Brad, go 1st. Thanks. Yeah, I, I agree going down the the path of deciding what.

78 "" (0) 00:21:45.089 --> 00:22:05.089 Terms of use are valid or not or should be valid or not. It's not there's not particularly helpful. I think we can decide as a group internally you know whether or not one thing provels over the other without reference to as a matter of law. I think it's as a matter of this group, we could decide that. It doesn't seem like that's probably where consensus is headed, but.

79 "" (0) 00:22:05.089 --> 00:22:16.139 That at least seems to me to be a plausible way of, of, of approaching this in terms of how we want the framework to, to, to, to work. My concern is that.

80 "" (0) 00:22:16.139 --> 00:22:36.139 For publishers and I think anyone who wants to govern how humans and bots interact with their websites, the only way to do that with any degree of granularity in a world where you might have some recourse if someone does something that is harmful, that is contrary to how you would like them to behave in using your content is through terms of use. These preferences.

81 "" (0) 00:22:36.139 --> 00:22:47.849 We're never going to suffice for that. So I agree that they are completely different things, and in that sense, the instrument that publishers and, and web operators use to.

82 "" (0) 00:22:47.849 --> 00:23:07.849 Government how parts some people interact with their sites needs to be preserved in its, in its full integrity or I don't think there will be any incentive. In fact, there will be a massive disincentive for anyone to use these preferences if there's a risk that doing so eviscerates what they have carefully drafted in terms of, in terms of terms of use. Whether you agree they should be valid or not, that's gonna be.

83 "" (0) 00:23:07.849 --> 00:23:26.399 I think the the the inclination from a from a publisher or web operator side and I think that that's, that that's fair. You want to put something out into the internet and you have to do it in a very naked way. And I think terms of use terms of service are pretty much, you know, the thinnished armor that you could potentially use to protect yourself.

84 "" (0) 00:23:26.399 --> 00:23:46.399 We're also not in control of the nature of the bots and human activity that, that, that come across our sites. Many bots don't identify themselves, many spoof their identity. Many maybe doing things that are nefarious, and I don't think that it's fair that our preferences which are supposed to be to the world, to any bot that comes to.

85 "" (0) 00:23:46.399 --> 00:23:53.699 Should just now kind of have some degree of force in the event that someone is doing something that they shouldn't be doing.

86 "" (0) 00:23:53.699 --> 00:24:09.029 And that they're not a good actor. And so I think in that case also, again, terms of service really need to be, protected in the integrity needs to be protected. I think there's different ways that we can address and describe this.

87 "" (0) 00:24:09.029 --> 00:24:27.449 You know, I think referencing as a matter of law may may sounds like it's too far for this room. I think that, you know, addressing the fact that one may prevail or the other is a choice that we could make or certainly ensuring that at the very.

88 "" (0) 00:24:27.449 --> 00:24:47.449 Very least that there's an agnostic approach to there is nothing in this that intends for cause I'm hearing a lot of things said that I think does sound to me like people do intend for preferences to govern and that the bots must always be able to rely fully on 100 % or whatever a very blunt.

89 "" (0) 00:24:47.449 --> 00:25:04.529 Very broad, very native preferences, and I don't think that that's going to be great for whatever it is that we produce here if that's the basis of how our preferences work. Thank you. Meredith Glenn and then we follow the queue. So what made you think about this? Maybe is that.

90 "" (0) 00:25:04.529 --> 00:25:19.649 Terms as someone over here says, the terms of use and govern human behavior and to the extent that they are treated as like quasi contract, they're between people, right? Like a bot can't sign a contract, I think right now.

91 "" (0) 00:25:19.649 --> 00:25:37.499 And so isn't the whole theory of the AI prepping that this is how the like technological interaction happens? And these are really separate like domains of one is like the legal purpose like the purpose of what you're doing. Like, are you using this information for what like.

92 "" (0) 00:25:37.499 --> 00:25:52.919 Human purpose, and that's the terms of youth question. But the question of like how your technological interaction with the website goes, that's an AI perhaps question. So to me, I think you can leave this as saying that they the perhaps don't.

93 "" (0) 00:25:52.919 --> 00:26:10.049 Override the terms of use and then just stop after the 1st clause of that sentence and let these be separate domains. The human purpose of the interaction with the website, which is governed by terms of use on the technological manner of interaction with the website governed by. Thank you, Len.

94 "" (0) 00:26:10.049 --> 00:26:29.309 So I have a couple of comments. The 1st is that I don't think the ITF in a technical specification has the remit to weigh into what legal contracts trusted an order applies. It just, it just isn't.

95 "" (0) 00:26:29.309 --> 00:26:48.389 Even if it did, let's say that we we put this in. Ultimately, whatever we write as a technical specification is going to be taken by an engineer embodied a piece of software and then stood up inside of an authoring tool or on a website itself.

96 "" (0) 00:26:48.389 --> 00:27:06.029 So how to set these preferences. The person stating those preferences, the person who is the publisher of the content may never ever be exposed to the RFC that is published by the ITF that has this technical specification in it. So they have NO connection or even awareness that.

97 "" (0) 00:27:06.029 --> 00:27:21.329 This is if it wasn't in there, that this is in there, that they are agreeing to this pressed and order. So I I think it's a it's a non starter. Thank you. I and and one additional comment about this. In the model as we're exploring it.

98 "" (0) 00:27:21.329 --> 00:27:41.329 There are different parties involved. The website operator may have the terms of service for the use of the website. The content creator, the content owners that are publishing through those websites, they may not be agreeing necessarily to all of this stuff and and we always should I think keep that in mind that we have.

99 "" (0) 00:27:41.329 --> 00:27:48.059 The age of large aggregators that take content from a variety of sources. So for my own use case.

100 "" (0) 00:27:48.059 --> 00:28:03.149 We publish trailers from our movies on a variety of video platforms. You know, it isn't just one platform, it isn't just one terms of service. We would like to if this AI preferences becomes a thing.

101 "" (0) 00:28:03.149 --> 00:28:19.199 Have the ability for our pieces of content to say, here are the purposes for these trailers and have a consistent set of preferences that actually go into all of the platforms that we publish those trailers through. So the assumption that the the website operator owns the entire.

102 "" (0) 00:28:19.199 --> 00:28:36.239 Body of content that's on that website and is publishing it and is setting all the rules for it, I think it's an incorrect model. And so here's my second comment. Evin? I think my understanding here is.

103 "" (0) 00:28:36.239 --> 00:28:51.869 The entire point of this exercise has actually been in deference to publishers to give them a tool to signal their preferences. So if, if you want the terms of service and the signals to be aligned.

104 "" (0) 00:28:51.869 --> 00:29:07.019 Then you should align them or don't state a preference. But if they're not aligned with your terms of service or your actual preferences, that's on you because I mean this exercises point which if NO machine can rely on your stated preference.

105 "" (0) 00:29:07.019 --> 00:29:27.019 Which is the entire point of the state? Well, I think I I honestly think to some degree what Brad said, that this is difficult to reconcile.

106 "" (0) 00:29:27.019 --> 00:29:35.159 Is exactly what we try to achieve with the original language. Scroll down a little bit and show the original language. We are essentially.

107 "" (0) 00:29:35.159 --> 00:29:51.539 We are essentially trying to express that there are NO elses related legal obligations in either case expressing preferences without prejudice to applicable laws including the terms of use are based on.

108 "" (0) 00:29:51.539 --> 00:30:08.669 Laws, right? Like they are sort of a, a instrument that operates within the framework of the law to express certain things that you can express because you have certain rights. Those are limited. Sometimes terms of use will exceed what you can actually express.

109 "" (0) 00:30:08.669 --> 00:30:28.669 But we are essentially saying like the preferences don't override any sort of proper exercise of your rights under the law that will include in many many cases well crafted terms of use. It will not include sort of terms of use that impose things that cannot be imposed as a matter of gone.

110 "" (0) 00:30:28.669 --> 00:30:37.559 And I think the the language is abstract and sort of light touch, but like it is that that paragraph is designed to.

111 "" (0) 00:30:37.559 --> 00:30:53.099 Ensure that these two types of instruments which speak to different audiences, as Meredith said, can exist and in the end that will not be us to determine what the relationship is here that will be sort of up to sort of if legal conflicts arise.

112 "" (0) 00:30:53.099 --> 00:31:13.099 To the legal system to make the determination based on what the legal basis for that is. And I, I think the current language is fine, sort of like it tries really not to change the status quo in that regard. And as like I think numerous people have said like that's also outside of the remedy of us to reorder sort of like types of legal.

113 "" (0) 00:31:13.099 --> 00:31:15.659 Arrangements as a work.

114 "" (0) 00:31:15.659 --> 00:31:33.269 Thank you. Thanks for Max. Max, are you gonna do a summary of like the things? I see the queues continued. Would you want to pup yourself to the end or ok Nick, go ahead. So on the point of publishers needing to.

115 "" (0) 00:31:33.269 --> 00:31:49.499 You know, express in the preferences something that's consistent with the terms of service. That's only possible if the set of preferences that we come up with allows for a full expression of the preferences that we have. And I am very concerned that we're steering towards a draft that is so minimal.

116 "" (0) 00:31:49.499 --> 00:32:06.629 Just to try to get something done that it won't be expressive enough to capture that. So if that's, if that is the position we're gonna take, then we must have a sufficient amount of preferences to capture the full range of expressions that we want to hold. I, I think like ok so I might be like.

117 "" (0) 00:32:06.629 --> 00:32:24.569 Talking too much, but I think the Kevin's point was like conflicts, not really like you know my task, but I think it was this was specifically talking about conflicts something, so if this preference is able to express that, right? And he put something else in the task, it's like I think that's what Kevin was going for. It's up to you to ensure that they are internally consistent way.

118 "" (0) 00:32:24.569 --> 00:32:41.249 But if the preference doesn't allow me, like if the preference doesn't express the thing or if I don't understand what the preference express has worked, not a category for grounding, and that is what is really, really important to me. And then, but yeah, I select one of the categories that does exist.

119 "" (0) 00:32:41.249 --> 00:33:01.249 Then my terms aren't, you know, like that's the problem is, is that fundamentally terms are a far more expressive way of expressing yourself, and we are currently steering towards a conclusion that is very limited in the amount of expressions that we're going to be made available. I just want everyone to be cognizant of that.

120 "" (0) 00:33:01.249 --> 00:33:21.029 Particularly as we go into like the use category there. Okay, thanks. Chris, I think I think Nate's actually just said what I wanted to say like we we we I mean the the the line of thinking with the terms of service is that.

121 "" (0) 00:33:21.029 --> 00:33:41.029 We need a much more expressive vocabulary that publishers can then use to express what they currently do through written in terms of service, but to to put it into a machine readable format. And I think if, if that's what ultimately we need, then, you know, that, you know, that's really, I think perhaps what we should.

122 "" (0) 00:33:41.029 --> 00:33:48.029 Focus our attention on, on, you know, on trying to build. The, the other point I wanted to make was.

123 "" (0) 00:33:48.029 --> 00:34:08.029 Yeah, this distinction between what terms of service for humans and you know robots.text is, you know, for bots. I I just wanna point out that, you know, bots are operated by humans ultimately and so they, you know, the humans are are ultimately sort of responsible for the bot behavior.

124 "" (0) 00:34:08.029 --> 00:34:23.489 So it it seems to me that it's, you know, it's a reasonable thing for us to try to do, which is, you know, align what we can express through written terms of service, but in in into sort of machine readable form. I I don't think it's acceptable that.

125 "" (0) 00:34:23.489 --> 00:34:40.049 That we have a situation where the the preferences that we can express through the vocabulary that we have at the moment just isn't, you know, isn't expressive enough.

126 "" (0) 00:34:40.049 --> 00:35:00.049 Justin? I don't think NO matter how many preferences do we offer, will we ever cover everything that everyone wants, someone's going to come up with a new one that they want. So I don't think we'll ever cover every use case. As soon as you apply terms of service as an option, then we might as well have NO preferences cause you can just put it on.

127 "" (0) 00:35:00.049 --> 00:35:11.879 In the services point to that. We don't need any preferences at that point, but we want it to be machine machine readable. I think that middle ground could be somewhere, here are the preferences the AMP preferences offers, there's a list that.

128 "" (0) 00:35:11.879 --> 00:35:31.879 It's created and then there is one additional preference that says, by the way, there's a terms of service over here. Hey mr. Carler I should probably go and read it because there's other stuff in it. But I don't think we'll ever agree on all the preferences because there's probably that we won't pick up. Thank you. Glenn, and let me apologize, I'm putting my hand up because my internet.

129 "" (0) 00:35:31.879 --> 00:35:35.879 It's not working it's been joining the queue. You know.

130 "" (0) 00:35:35.879 --> 00:35:55.879 I want to jump in and say show my support for this notion that we have a problem if we limit you to the terms the the preferences we provide. We do not know all the use cases that may come out after the specification is is published, so it's very likely new things will emerge. We do not yet have a protocol that.

131 "" (0) 00:35:55.879 --> 00:36:12.149 Is getting steamed that would have extensible preferences so therefore we're kind of limited to whatever we produced. So it's very likely that whatever we do produce on the current path will have limitations and will eventually have to have some way either through service, other expressions.

132 "" (0) 00:36:12.149 --> 00:36:28.379 Ways to express new use cases and new preferences. Recognizing that, I don't think we should put a lot of effort into specifying that this is it, this is the only way to do it. I think we should fall back to a simpler time in this process where we said.

133 "" (0) 00:36:28.379 --> 00:36:48.379 The preferences are a way to express it. They're an optional way. They are an additional way. You may have others, they're not the exclusive way. It is a, a, it is a tool that you can use. We offer it up as that. If you choose to use it, go forth. If you choose not to use it, that's fine too. There's other ways. We have CGPA.

134 "" (0) 00:36:48.379 --> 00:36:59.399 We have other ways that are probably in the industry, so we should recognize that and and and pull back a little bit from trying to express it as if we are writing the final word on the subject cause we're not.

135 "" (0) 00:36:59.399 --> 00:37:17.399 Good, thanks Paul. Yeah, I think I, I get the impression that some people are arguing at the moment that sort of like the draft as it exists right now would somehow somehow state that preferences override anything. It does.

136 "" (0) 00:37:17.399 --> 00:37:37.399 Right, like it does carefully try to say like these exist, they are specified here, they are meant to be like this, and they do not overwrite existing legal arrangements. So like I sort of like this is not a, this this feels bit like an argument should the preferences overwrite the terms of youth or the other way around?

137 "" (0) 00:37:37.399 --> 00:37:56.369 We're actually in the current state are not saying that something takes preference over something else and I think that is the only way to resolve this by being neutral with regards here and deferring this really to the level of the law. Thank you guys agree with that. I'm Brad.

138 "" (0) 00:37:56.369 --> 00:38:11.880 Yeah, so, just to throw, most possibilities into the mix, but I hope maybe some maybe in how we might address this. I added some language in issue number one.

139 "" (0) 00:38:11.880 --> 00:38:29.730 Okay, sorry. I'm not aware of hearing it's an outside really loud to myself, so it's like, sorry, should be the opposite problem. So, one 60, issue one 60 is I added a comment too with language.

140 "" (0) 00:38:29.730 --> 00:38:45.240 That I think encapsulates some of the, the, what i've, what I've heard right now essentially striving for neutrality, an agnostic approach, but do not override principle referencing both agreements, existing law.

141 "" (0) 00:38:45.240 --> 00:39:04.200 And this is not a specific position, but it's a an attempt to be, to be helpful. And maybe I I don't know if we can put it on the screen. It's, one 60, it's one 60. Yeah, I put it in the chat themselves create rights or prohibitions.

142 "" (0) 00:39:04.200 --> 00:39:24.200 Sorry? Inspecting preferences. I just dumped it on the chat. Any. Is it 16160 Inspecting preferences. Yes. I put the link on the screen. So we're we're just about to get to, to this issue.

143 "" (0) 00:39:24.200 --> 00:39:40.440 Yeah please do tie together yeah yeah but sure so we can look at it in another context, but NO thanks. Thank you for that. That's helpful. Max. Yeah. Go. All right, there was a lot.

144 "" (0) 00:39:40.440 --> 00:39:57.870 There was said, so I don't think I'm gonna be able to touch all on every single point and there was some cross talk that I think diminished the need to answer some things and improve the need to answer others to to to Mike's points on.

145 "" (0) 00:39:57.870 --> 00:40:17.870 One is for humans, one is for bots. I think that would be a very simplifying thing to do, but I think it's completely incorrect unfortunately. I, I understand the instinct, but it's, it's just not true. Like terms of use apply to all visitors to the site.

146 "" (0) 00:40:17.870 --> 00:40:35.100 Can be written that way, and often are, at least in the examples that I'm aware of, they're written as for all visitors, they're intended to capture both human and bot behavior, and I think Chris Needam's point also like bots are runs by human. They're not making all decisions on their own, any responsible.

147 "" (0) 00:40:35.100 --> 00:40:52.110 The owner of a bot has human decision making in how they construe it. None of these decisions are so entirely devoid of of of human interaction is to be completely and irrevocably different. So I I.

148 "" (0) 00:40:52.110 --> 00:41:07.980 Pretty much declined the premise that there is a bot document and a human document. In terms of use is specifically all encompassing from from my perspective, and in terms apply to all users regardless of of of origin.

149 "" (0) 00:41:07.980 --> 00:41:25.980 And on machine readability, I just want to put one thing in, I know we'll, we'll do a lot of like the we're a little bit touchy on legal things, I understand, but I do think it's worth noting that the EU copywright directive explicitly puts out that terms of service are machine readable.

150 "" (0) 00:41:25.980 --> 00:41:41.880 That's a, that's a data point that I think we should take pretty seriously. Again, there's a lot of other things that, that got cross referenced. I think there's some interest in moving up to one 60 as well, so we'll talk more on on list, I'm sure. But I I think.

151 "" (0) 00:41:41.880 --> 00:42:01.880 I'm not hearing really anything substantive from why terms of service are not an agreement that that matters and I think that's sort of the initial question I posed and I didn't really hear a counter to that. So I listened to this discussion and it seems like there are.

152 "" (0) 00:42:01.880 --> 00:42:22.190 Number of different forms of pushback which you heard, so I'd encourage you to go and and keep on developing your proposal if you if you want to. What I heard that was interesting to me was especially what Meredith said about the difference between, you know, it's one thing to say, yeah, I have a preference that you.

153 "" (0) 00:42:22.190 --> 00:42:25.260 Can but versus you can't.

154 "" (0) 00:42:25.260 --> 00:42:45.260 And less, you know, it's the additive versus subtractive kind of discussion because in in in it it's very common pattern in, in internet protocols that the default is safe and then the extensions can take away from that. Well, if we're gonna talk about defaults and additives, then we need to talk about the top line categories.

155 "" (0) 00:42:45.260 --> 00:43:03.360 Because I understand, then you're automatically saying your default has add additions automatically regardless. And if that's, if that's the line we're taking, then we have to talk about the top line category again. Well, that's not the line we're taking, that's my observation. I understand what I'm saying.

156 "" (0) 00:43:03.360 --> 00:43:19.980 So what I would encourage you to do is, I I also heard an interesting suggestion was, I think it was from Paul that this might make more sense as an attachment robots.tech specific addition rather than a vocabulary term.

157 "" (0) 00:43:19.980 --> 00:43:38.160 Because it does seem like it's specific to robots. I don't believe it is, but we can talk more about that later. At the end I'm I'm looking for ways that we can get you what you want in a way that gets consensus, and and that those seemed like the most promising ways you could make modifications to.

158 "" (0) 00:43:38.160 --> 00:43:54.750 Just get it towards which one to make. And I think like to to Paul's point, so is there anything that you think is in the draft that contradicts what you're saying? Like forgetting about being very specific about it, but do you think there's something in the draft?

159 "" (0) 00:43:54.750 --> 00:44:14.750 That says like the opposite or implies the opposite because I, I I like Paul, like Paul said, I don't see there's anything in the draft that does it like so you want to be more specific, but I was just wondering if there's anything contradictory in the draft. Yeah we look I I outlined it a bit in in the issue and there's there was a healthy preamble in there before the actual like host language.

160 "" (0) 00:44:14.750 --> 00:44:20.100 So I'd i'd refer people back to that really. I do think that the, the language is not.

161 "" (0) 00:44:20.100 --> 00:44:35.730 Perfect, that's why I proposed new language, but again, like I think if people want to like take a look at at the issue and and we can go from there, I think that's perfectly fine. Yeah I think I I don't think that our intent is captured in the current drive. All right, thank you.

162 "" (0) 00:44:35.730 --> 00:44:53.610 When you say our, I mean myself. Okay. The royal B I use the royal Wee a lot and people get confused. I'm sorry. Now I just that English. All right, see yourself as royal. It's ok. I did look.

163 "" (0) 00:44:53.610 --> 00:45:12.180 English is not my 1st language. I'm sorry about that. I grew up in a Russian speaking people. The more you know. So should we move on to one 60? Is that enough discussion of that one for now?

164 "" (0) 00:45:12.180 --> 00:45:27.990 So, there's still contention about the text in three.two. Alyssa started this issue made some specific.

165 "" (0) 00:45:27.990 --> 00:45:47.990 Requests and pointed out some issues and she suggested significantly pairing back the text starting from use to be ignored. So that provides less spotter for future debate about what is really intended here. And then she had a suggestion about a modification.

166 "" (0) 00:45:47.990 --> 00:46:06.840 To try and and capture the original intent that Lyla had, which we can read perhaps we should refresh ourselves on what three.two contains 1st, if the internet allows us to.

167 "" (0) 00:46:06.840 --> 00:46:22.860 Okay so this is the applicability and effect section and if you recall we added.

168 "" (0) 00:46:22.860 --> 00:46:40.830 This list here, we've we've edited it and gone through it a few times. There's still some questions about it. I think what Liz is proposing is, yeah.

169 "" (0) 00:46:40.830 --> 00:47:00.830 This block of text is what she's focusing on mostly. Whereas this up here I don't think she has as much comment about. So her proposal was parties must might choose not to honor preferences they receive on the basis of expressing.

170 "" (0) 00:47:00.830 --> 00:47:04.140 Payments that have with other parties, which we've discussed right now.

171 "" (0) 00:47:04.140 --> 00:47:24.140 For legal compliance purposes or for other reasons. Historically entities engaged in automated processing of digital assets and supportive safety education, scholarship, research, preservation, interoperability, accessibility and other parties have needed to make such choices in response to existing preference signaling mechanisms and effective parties have had to grapple with the consequences. The preference.

172 "" (0) 00:47:24.140 --> 00:47:43.740 Statements describing this specification are NO different. One of the merits of of what Alyssa is proposing here is we're, we're very carefully talking about or making an observation about what has happened in the past or what people do rather than making normative statements or or trying to steer the outcomes.

173 "" (0) 00:47:43.740 --> 00:48:02.520 And then Brad just 11 min ago. Brad, do you want to walk us through your proposal? Sure, so it's I guess, dresses or tries to kill two birds with one.

174 "" (0) 00:48:02.520 --> 00:48:22.520 So are you suggesting a wholesale replacement of the section or so it's not an addition and it doesn't necessarily preclude including what Alyssa has suggested. But I I think what we wanted to do was really pair back, the language so that it I think how.

175 "" (0) 00:48:22.520 --> 00:48:27.270 Problem we've had in the past is that we're starting to trip over our own.

176 "" (0) 00:48:27.270 --> 00:48:47.270 Feet a little bit when it comes to being very careful about saying what you might decide to do or not do in a way that I think is intended to satisfy those who really would like it to say you don't have to follow it in these circumstances, but I think that we're by taking the step and kind of identify.

177 "" (0) 00:48:47.270 --> 00:48:55.740 Certain use cases where you could decide not to do something, not to, not to respect the preferences. It it there's a, there's a strong.

178 "" (0) 00:48:55.740 --> 00:49:12.060 Implication of endorsement of, of that approach. I think the way that Alyssa has put it is kind of, more neutral and that it acknowledges what has, you know, kind of what has already been happening in the past without necessarily taking a position on.

179 "" (0) 00:49:12.060 --> 00:49:28.830 How that needs to pan out and that this approach really sits in that world and is NO different from, from others. And so our my our well we, my suggestion is.

180 "" (0) 00:49:28.830 --> 00:49:44.160 That we we stay completely neutral, that these are, we we clarified, you know, overly over clarified that there's NO enforcement mechanism and that we use the word that is agnostic.

181 "" (0) 00:49:44.160 --> 00:50:04.160 As to conformance that my parties doing data processing. So we're offering these preferences and we're saying what you decide to do in terms of recognizing them or not is not going to be a direction from, from, from the specification and that things that.

182 "" (0) 00:50:04.160 --> 00:50:24.160 Might influence how you decide or not decide to respect lie outside of the, the world of the specification, including legal rules, including contractual rules or or or agreements which which might exist. And then there's some of the same language that has been included there, which we've seen in.

183 "" (0) 00:50:24.160 --> 00:50:31.290 In in the past I think, you know, it it probably kind of hits the same points in a few different ways, but that is the.

184 "" (0) 00:50:31.290 --> 00:50:49.620 That is the the the the goal points. So before we get into discussion here, I'm wondering if it would be beneficial to take a break because it's about that time and also because we have a lot of text that is being introduced here and it might be good to give people a little bit of a chance to read before we.

185 "" (0) 00:50:49.620 --> 00:51:09.620 Get into it too deeply. So how about we take, let's say a 15 20 min break? Yeah. Just like Meredith, did you wanna make a quick statement like get a hand up or you can wait. I can do it. Okay, that's good. That'll let us just launch into the discussion when we get back. So let's come back at, yeah, about 1045, and.

186 "" (0) 00:51:09.620 --> 00:51:14.430 During the break, please familiarize yourself with, the original texts.

187 "" (0) 00:51:14.430 --> 00:51:34.430 Alyssa's proposal and now Brett's proposal as well. Mark, could we put that up? I don't have internet. Oh, so so neither of the internet connections work for you? The super flicky. I mean, it will work for like 2 s and going Both both of them. Yeah. Come on. What who did you.

188 "" (0) 00:51:34.430 --> 00:51:40.560 I, I'm I'm following because we're at Cisco. Really?

189 "" (0) 00:51:40.560 --> 00:52:00.560 Doesn't work here. Yeah, what else do we have? Yeah. Okay well what we'll do is I'll i'll.

190 "" (0) 01:13:53.490 --> 01:14:09.360 But in this case we're talking about relocation. So.

191 "" (0) 01:14:25.500 --> 01:14:45.500 Because yesterday. Windows, right? I apologized. I've noticed. Yep.

192 "" (0) 01:14:45.500 --> 01:15:03.210 Yes.

193 "" (0) 01:15:23.210 --> 01:15:32.130 Worked out.

194 "" (0) 01:15:32.130 --> 01:15:52.130 Okay, sorry.

195 "" (0) 01:17:32.130 --> 01:17:52.130 Your email address followed by two and it might be one for me. Okay.

196 "" (0) 01:17:52.130 --> 01:18:12.130 What I love is is that they make you put the badge on your chest and then you have to refer to it to get your password. I'm not gonna give you any. Okay, so let's.

197 "" (0) 01:18:12.130 --> 01:18:15.420 Get started again.

198 "" (0) 01:18:15.420 --> 01:18:33.420 We're on, 1902 which is the introduction. Oh wait, NO, sorry, we're not. 6116. Let me find that. I do have a pull request out for Brad's text.

199 "" (0) 01:18:33.420 --> 01:18:53.420 Stop doing that. I also have a pull request up that just rewords the statement that has May in it, not Addressing's secondary. So one of the things I'll observe is we're, we're getting in the habit of of putting up text and battling text versus other texts and.

200 "" (0) 01:18:53.420 --> 01:19:13.420 And and that doesn't help us move forward as much as talking about the underlying issues in the text that we have. And so I I'd like us to focus more on the issues that we have than the specific proposals at least to start with because that'll give us a better idea of what proposals might actually move us forward. So the, the issues that.

201 "" (0) 01:19:13.420 --> 01:19:31.890 What Melissa brings up are especially I think here, the, the text, the, the detailed text about when use preferences can be ignored, provide spotter for future debate about what is really intended, so if you look at the current text in the spec.

202 "" (0) 01:19:31.890 --> 01:19:48.960 She's talking about from here forward and this has come up for in discussion. Someone unmuted? Mike, are you on muted?

203 "" (0) 01:19:48.960 --> 01:20:08.960 I think it's ICMP, so let me ok so just keep that in mind, I don't want this to become, you know, just a battle of of which proposal we want to to advocate for, but more trying to address the underlying issues.

204 "" (0) 01:20:08.960 --> 01:20:27.900 So Meredith before we left you, you had, I was just gonna strongly support Alyssa's general approach as one that doesn't get us into a fight about a specific list of, forward looking.

205 "" (0) 01:20:27.900 --> 01:20:46.860 Permitted reasons to ignore this, but through a set of examples does give less sort of bold or legally sophisticated users some idea of what we're talking about here. I worry that.

206 "" (0) 01:20:46.860 --> 01:21:04.110 Doesn't list any of those specific historical examples is gonna be a hard signpost for people who don't know what this is talking about. And then I think most pragmatically thinking about the disability community that the people.

207 "" (0) 01:21:04.110 --> 01:21:23.910 Like that the structural cost to them is very high if this isn't clear, and so here having some signpost of these of what is being attended to be sort of carved out is really important from both a user standpoint and an equity standpoint.

208 "" (0) 01:21:23.910 --> 01:21:43.620 Thank you. So, I, so, so my view on this is that the, that list of examples that, that, that are currently in the draft under, you know, what, where, where preferences maybe ignored.

209 "" (0) 01:21:43.620 --> 01:22:01.290 Really has NO place in this document, like what this is trying to do, and I think is sort of bring some of the, the sort of surrounding legal context into our draft, and I think that's not not the place that we should be. I think where we should be is focusing on.

210 "" (0) 01:22:01.290 --> 01:22:21.290 The preferences that can be expressed and leaving the cases where organizations and people have have rights to, you know, to use content regardless of preferences, like let's leave that to legislators that this this all of this I see is, has really sort of beyond being.

211 "" (0) 01:22:21.290 --> 01:22:22.350 Beyond.

212 "" (0) 01:22:22.350 --> 01:22:41.460 The remit of our group. So I I, my, my view is that the, the 1st paragraph that you've highlighted here is fine, the rest of it should be struck. So Chris, Alyssa's proposal was to replace all that highlighted text with, with this.

213 "" (0) 01:22:41.460 --> 01:23:00.900 Does that yeah NO because it's still pointing to the same thing. The fact that it says historically sort of just lends ambiguity I think into what we're trying to say. I think it's better that we say nothing than we say, than we sort of point to particular examples.

214 "" (0) 01:23:00.900 --> 01:23:20.310 Lila? Hello, so for you know given that I wrote sort of the initial thing, I think it's pretty obvious that I I support some version of this sort of.

215 "" (0) 01:23:20.310 --> 01:23:40.310 Illustrative list being in the document itself. I strongly disagree with Chris's proposition that this is legal in nature. This is not. It is simply a set of rationales why diff.

216 "" (0) 01:23:40.310 --> 01:23:45.960 Think any kinds of users users might.

217 "" (0) 01:23:45.960 --> 01:24:05.960 Choose not to, you know, I I guess my point is like if it's ok for publishers to override things with their terms of use, I think we need to be clear that, like, even if you don't aren't in a position to have like your own terms of use and you're asserting things, like, you know.

218 "" (0) 01:24:05.960 --> 01:24:25.230 There are other rationales why these things might not be followed and this is meant to be illustrative, normative, not at all legal, and I don't think like Alyssa's text here.

219 "" (0) 01:24:25.230 --> 01:24:41.550 Adds the term for legal compliance purposes, which I actually think is unhelpful, so that would be something that I don't think should be here but having some kind of illustrative list for.

220 "" (0) 01:24:41.550 --> 01:25:01.550 Yet the non technical, which is the vast majority of all internet users I, I think is, is helpful here. And the other reason I think this is important is to the extent that, this specification does end up in some journey.

221 "" (0) 01:25:01.550 --> 01:25:03.930 Restrictions being.

222 "" (0) 01:25:03.930 --> 01:25:23.930 In some way legally enforceable, I think it is important that the specification itself is clear that there are rationales and justifications for ignoring it so that we don't end up in a situation where somebody who is just trying to.

223 "" (0) 01:25:23.930 --> 01:25:43.940 You know, make a website accessible for themselves, might have some sort of like legal problem. So, so Leela, if I could, just to make sure I understand, it sounds like you prefer the current text to remain, but it sounds like you'd be amenable to changes along the line that.

224 "" (0) 01:25:43.940 --> 01:25:50.250 Alyssa or something like that, you know, with some modifications, but you'd be against just removing that test.

225 "" (0) 01:25:50.250 --> 01:26:10.250 Yeah, exactly. So, so I, the reason I think the illustrative list is useful it gets to what Meredith was saying, about a list because that can become, you know, the ceiling and say like these are the only things that are allowed. So.

226 "" (0) 01:26:10.250 --> 01:26:16.440 What I like about what Alyssa has done here, has made it.

227 "" (0) 01:26:16.440 --> 01:26:33.720 A little more like high level and not concrete and I think that that has trade offs, and I'll just say that I, you know, I hope we get to Tim at Robot's proposal about like.

228 "" (0) 01:26:33.720 --> 01:26:50.100 He has he has a you know a strong or sorry, they, pardon me, have a, I have a strong, you know, idea there of like, well, there should be some things that are very concretely carved out for public interest.

229 "" (0) 01:26:50.100 --> 01:27:06.600 Purposes, and I have the same concern with that list because it becomes this, you know, oh, these are the only things that are allowed, which I don't, I don't know that that's what we want to do. So I am amenable to changes being made.

230 "" (0) 01:27:06.600 --> 01:27:22.530 And further discussion on the question of whether to include, you know, a list of examples or an explicit set of exclusions, but I, I.

231 "" (0) 01:27:22.530 --> 01:27:42.530 Yeah, I I strongly believe that some referenced even like non legal mechanisms like a terms of service, being a justifiable rationale for ignoring the preferences is important to effectuate you know of.

232 "" (0) 01:27:42.530 --> 01:27:45.210 Kinds of public interest.

233 "" (0) 01:27:45.210 --> 01:28:04.290 Issues that are otherwise left outside of it, and if that's completely left outside of it, then it feels like especially risk averse institutions like universities, libraries, cultural heritage institutions will interpret it as being.

234 "" (0) 01:28:04.290 --> 01:28:20.940 This is the final word and there are NO exceptions, and I really worry about what that impact, the downstream impact of that would be. Thank you. Timmed Robot? Tim androbot.

235 "" (0) 01:28:20.940 --> 01:28:36.360 I just wanna remind us that during discussions of our two existing categories, both of our efforts in those included attempts to protect.

236 "" (0) 01:28:36.360 --> 01:28:56.360 Public interest uses, like the, the discussion about whether to include like translation or transcription or whatever in search, and so the idea that these, either the examples or in my draft, the exclusions.

237 "" (0) 01:28:56.360 --> 01:29:13.410 I don't have a place we've already like the they're already here in the room with us as it were. And so I think that, I think the giving concrete examples is, is helpful.

238 "" (0) 01:29:13.410 --> 01:29:33.410 I respect the idea that they should be especially when they're kind of in like like in the existing text that they're not described in the way as to imply that they are the only examples, but that establishing them at a kind of a higher level could make the rest of our work.

239 "" (0) 01:29:33.410 --> 01:29:53.070 Much easier, as we don't have to, you know, if they're protected at a high level, we don't have to make each category try to protect them. Do you do you want to talk to your draft now? Sure. So the, my draft is linked from.

240 "" (0) 01:29:53.070 --> 01:30:09.750 Ticket 160I also want to call attention to the fact that linked from the introduction or at the bottom, and the informative references, there's a diff. Look at that. Why am I seeing different things on the two? Oh, there it is.

241 "" (0) 01:30:09.750 --> 01:30:29.580 It's just catching up. No, ok. Yeah, that's slowing. Yeah, I fine. Okay. All right guys. Do you wanna look at the diff or or this document? I think people can look at the diff to kind of but basically.

242 "" (0) 01:30:29.580 --> 01:30:44.910 Well if we if we want to look at the diff quickly 1st and then we'll come back to this text, because I think that diff will help people kind of understand the proposal.

243 "" (0) 01:30:44.910 --> 01:31:01.650 And so and 6.2 is the actual link to the diff. So you're just adding that section is that the one? It's on my screen, it's just not on that screen. Okay. So it's it's basically it's adding.

244 "" (0) 01:31:01.650 --> 01:31:19.320 It's adding a new 3.2 section and bumping the applicable an effect to 3.3, and it's it removes the examples.

245 "" (0) 01:31:19.320 --> 01:31:39.320 That from applicability and effect as they are largely, they're very similar to the exclusions that are carved out in the new section 3.2. Oh, I see, ok. That's really hard to process.

246 "" (0) 01:31:40.830 --> 01:32:00.830 Okay, so you're just taking them out of there and put them into these section effectively? Yes, and also giving them the weight of exclusions versus examples of potential reasons to ignore them.

247 "" (0) 01:32:00.830 --> 01:32:16.919 And they're further kind of restricted in the 3.2 section, so e.g., our attempt, my attempt to kind of limit.

248 "" (0) 01:32:16.919 --> 01:32:35.579 Internationalization and localization so that it's not creating a an opportunity for kind of exploitation from the perspective of content owners was to limit it to end users.

249 "" (0) 01:32:35.579 --> 01:32:55.579 So, underlying I think a lot of the discussion we've had about this, this section has been this tension between just doing illustrative examples to note the relationship of our specification to things outside of our scope and authority versus making normative statements about what is in.

250 "" (0) 01:32:55.579 --> 01:33:04.859 Is not what what preferences do and do not apply to. And it sounds like you are leaning more towards making normative statements. Would you say that's true? Well, it's about.

251 "" (0) 01:33:04.859 --> 01:33:20.489 Kind of creating specific holes in the preferences so that the, so that we don't have to then try to create those kind of specific protections or holds in each category. Right.

252 "" (0) 01:33:20.489 --> 01:33:40.489 But but I I say that because, you know, one of the contentious issues has been, you know, when we pull those into our scope, we're effectively interpreting the law, in this area or or or try perhaps being seen as trying to preempt it. And, and that's I think why we.

253 "" (0) 01:33:40.489 --> 01:33:45.899 We ended up where we did where it was more just illustrative examples to kind of balance those concerns.

254 "" (0) 01:33:45.899 --> 01:34:01.739 So just noting where we've been. I wasn't, I wasn't trying to propose this as an expression of any legislative reality or jurisdiction anywhere. But from the perspective of.

255 "" (0) 01:34:01.739 --> 01:34:20.369 A healthy and robust internet given? Yeah, I, you know, obviously we're not making decisions in the room and I'm not making a decision on what I think in the room.

256 "" (0) 01:34:20.369 --> 01:34:36.329 But on 1st blush I I think what Alyssa is suggesting is pretty elegant in the sense of just just as I think we would be beating our head against the wall trying to come up.

257 "" (0) 01:34:36.329 --> 01:34:55.499 With preferences that accommodated every possible use case that publishers wanted to be able to express a preference on, I think trying to come up with an exhaustive list of questions would, would, would crash on the same shows and, and would also have the problem of not being future proof or flexible.

258 "" (0) 01:34:55.499 --> 01:35:11.969 And I think simply, and I and I agree, I disagree with the statement that that Alyssa's solution brings in legal issues. It expressfully doesn't, it is simply a historic statement of fact that helps express.

259 "" (0) 01:35:11.969 --> 01:35:28.469 A rather fuzzy norm that we want to be able to preserve. My only hesitation with Alyssa's language is the vague reference to suffering consequences, and I'm not sure why that's there or who it's for.

260 "" (0) 01:35:28.469 --> 01:35:48.469 For what consequences we're talking about. I think we're simply saying these are preferences. Preferences might be disregarded for reasons. Here are reasons that have historically, been a reason. You know, that, that, that has been I mean, we don't even know.

261 "" (0) 01:35:48.469 --> 01:35:50.579 Not say that has been accepted. It's just like.

262 "" (0) 01:35:50.579 --> 01:36:10.579 Historically, these are reasons that people have disregarded robots subtext. I think like Alissa's point was like people have had to suffer the consequences, not that they have to, right? Like affected parties, meaning the site that is being scraped? Everybody, all of them, right? Like I I think it's big enough, but it in the past.

263 "" (0) 01:36:10.579 --> 01:36:22.199 That's the important part is like it's putting the historical context and say like hey like people had to deal with whether the site owners or the callers had to deal with not obeying those preferences was like the point she was trying to make, I think.

264 "" (0) 01:36:22.199 --> 01:36:42.199 This is a bit of a caveat after I guess. Thanks Kevin, thank you. You ready? Yes, thank you very much. On the issue of ignoring preferences, I tend to agree with Chris that exclusions like this have NO place in this document. Granting exceptions or exclusion.

265 "" (0) 01:36:42.199 --> 01:37:02.199 Just to their preferences is not what the majority of implementers will expect the standard does, and certainly content owners will not want to do this. It is not a normative document as many say, and it cannot be used to determine when content maybe accessed or used despite valid preferences or even e.g. reservations of rights. There is NO risk of people being.

266 "" (0) 01:37:02.199 --> 01:37:22.199 Prevented from actions because again this is not an access control. We could live with a general reference to the fact that the law may provide for access under certain circumstances in certain jurisdictions, but don't support the inclusion of any concrete examples that content owners could be portrayed as subscribing to by dent of using the standard.

267 "" (0) 01:37:22.199 --> 01:37:42.409 Thank you very much. So, you know, just to clarify, you're talking about the the text in the draft the text Alice is proposing or text. Which one are you talking about? Or all of them? I was referring specifically to the text on a 3.2 that outlines the specific examples I do see.

268 "" (0) 01:37:43.499 --> 01:38:03.499 Proposal here, but in any case exclusions from preferences are not something that should be covered in our view here, given the importance that was described to the standard in a number of jurisdictions, in particular for determining things like reservations of rights, I don't think this has any space or any purpose beyond. It shouldn't.

269 "" (0) 01:38:03.499 --> 01:38:11.399 Not in any case determine what a content owner is granting access to by ways of exceptions or exclusions. Thank you.

270 "" (0) 01:38:11.399 --> 01:38:31.399 So, so I I hear you're you're you have a problem with the current text in the draft and it sounds like with Timmed robots proposal as well. The the text that's currently on the screen, this proposed modification are are are you more amenable to that or does that have the same problems in your mind? I mean it just offers.

271 "" (0) 01:38:31.399 --> 01:38:51.399 It provides for a lack of clarity. I mean if if we have historical examples as to why preferences could be ignored, there are many examples, historical examples of other things. I don't think this serves practical purposes, but it does provide for a potential confusion or indeed misrepresentation of the intention of the standard implementer.

272 "" (0) 01:38:51.399 --> 01:39:01.859 When expressing their preferences. So when people say NO scraping, that doesn't mean but yes scraping potentially for this list of activities.

273 "" (0) 01:39:01.859 --> 01:39:21.859 Yeah, I had a couple of questions for Lila on on what she said. So like looping back a little bit farther in the in the conversation. You said that to the extent that like these jurisdictions, there will be jurisdictions where this is legally enforceable, it's important that the.

274 "" (0) 01:39:21.859 --> 01:39:41.549 Specification is clear on the rationale for when it should be ignored, like when preferences should be ignored even in like a legal jurisdiction. So like that seems to make the implication that we should be able to decide what the limits to a law is, is that what you were intending to say or did I misunderstand there? And then the other thing is I didn't really understand.

275 "" (0) 01:39:41.549 --> 01:40:01.469 Your point on terms of use. So I was I was hoping you could clarify, it was sounded like what you wanted or what you were saying is that like we can make statements in this group about what terms of service can't prevent, and so I I wasn't sure if I was misunderstanding you or or or something like that, so I wanted to task on that as well.

276 "" (0) 01:40:01.469 --> 01:40:17.219 Alila, you can come off mute and speak if you wanted to. Sure, thanks. And so sorry who was the speaker, was it max? That's. Sorry. Hi, NO, that that's ok. Sure Okay so.

277 "" (0) 01:40:17.219 --> 01:40:35.939 Take, well I guess taking 2nd point 1st, which gets at this sort of issue of, like, these are preferences, they, I.

278 "" (0) 01:40:35.939 --> 01:40:55.939 As far as what I'm hearing I think we're all in agreement that they are preferences and they are not in and of themselves legally enforceable. There maybe a scenario, there maybe scenarios where jurisdictions could choose to make them so, but that like on their.

279 "" (0) 01:40:55.939 --> 01:41:04.379 Face they're not and so that's that's why my like I am in agreement with the people who are saying.

280 "" (0) 01:41:04.379 --> 01:41:24.379 Terms of use, could either expand upon or change or override, like if you're in a legal scenario where there's a legal dispute, it is my opinion that like, yeah, the legal thing would override this preference. Okay, so that's just like I.

281 "" (0) 01:41:24.379 --> 01:41:32.789 I was, I was referring to sort of the legal dispute scenario, and.

282 "" (0) 01:41:32.789 --> 01:41:52.789 What I am saying is that unfortunately, end users and a lot of public interests actors don't get to just sort of impose their own legal, you know, their, their legal terms on other people. We have, there's NO mechanism.

283 "" (0) 01:41:52.789 --> 01:42:10.709 By which to do that. And so to kind of have an even playing field between the kinds of like publishers and website hosts that, that have that legal mechanism to do that, I think that it is helpful for.

284 "" (0) 01:42:10.709 --> 01:42:30.709 The public interest actors who both want to be, who want to be in conformance, want to respect preferences, most of the time, but there may maybe scenarios where, it isn't in conflict with their.

285 "" (0) 01:42:30.709 --> 01:42:49.979 Or mission or maybe even their own legal obligations. E.g., right? Universities in the United States have obligations under the Americans with Disabilities Act to provide accessible formats for for their students, and so there maybe legal obligations or other reasons.

286 "" (0) 01:42:50.000 --> 01:43:07.339 And so I I think it is important and helpful to state that these preferences are not intended to override those sorts of legal rights or concerns either. So to in my mind.

287 "" (0) 01:43:07.339 --> 01:43:27.339 This is giving parity to, to the different sort of rights and expectations that people have, and not, and I think that there's for the people in the chat who are saying things like oh this could be misconstrued as like a license or anything like that.

288 "" (0) 01:43:27.339 --> 01:43:45.230 It seems to me that there, there is language we could put in here that could mitigate that concern, but, you know, again, I I am ok with even sort of this Alyssa's more pared down.

289 "" (0) 01:43:45.230 --> 01:44:04.580 List but I think it's important to have, a clear statement that even outside of the context of some sort of legal arrangement, there are valid and justifiable rationales for the preferences to not be followed.

290 "" (0) 01:44:07.700 --> 01:44:27.700 Thank you. Thanks, Max, did your questions get answered? Maybe not your satisfaction, but did they get answered? Sure. Okay, thanks. Brad? Thanks. I I'm in agreement with those that have expressed the concerns about having a list, whether they're.

291 "" (0) 01:44:27.700 --> 01:44:39.530 Mandatory or or just a suggestion of, of, of, instances where you might not respect the preferences. I.

292 "" (0) 01:44:39.530 --> 01:44:59.530 Don't know if in my own mind I am reconciled to whether or not if there were such a list, we would also not want to have a list of instances where you would be recommended to respect them, particularly in the context of the discussions that we've had in the room and the impact on creators and publishers and areas where they are very.

293 "" (0) 01:44:59.530 --> 01:45:01.010 Be likely to be.

294 "" (0) 01:45:01.010 --> 01:45:20.270 You know, where there's, there's likely to be harm if you don't, if you don't respect them. I I, I don't think that it's useful to go down that road. I think it's better to, to take a step back and to embrace more of a, a an a neutral approach which was.

295 "" (0) 01:45:20.270 --> 01:45:40.270 Really what my language that was posted just a few minutes ago was intended to, was intended to to address which which primarily the difference that that that that my language gets to is, is taking out the list and just reframing some of the wording to at.

296 "" (0) 01:45:40.270 --> 01:45:50.030 Get to a sort of sense of of neutrality or being agnostic about the rules and norms and frameworks that exist outside of this specification.

297 "" (0) 01:45:50.030 --> 01:46:10.030 I think what also concerns me about, some of the robots approach is that I'm not sure that if you were to follow them, then in all instances you might not actually risk incurring some liability in certain jurisdictions. I don't think we can know that for sure and certainly reading them on their face that general enough.

298 "" (0) 01:46:10.030 --> 01:46:29.600 To give some folks comfort that, oh, this seems to be fine thing to do and actually a worthy thing to do and I don't understand that would be a worthy thing to do, but I I I don't think we'd want to suggest any pathways for folks to get themselves into legal legal trouble and we're not really possessed of the knowledge and experience to know.

299 "" (0) 01:46:29.600 --> 01:46:46.430 For the jurisdictions that this is going to apply in, whether that's, whether that's the case, and I think that really does council again to come back to, we're not going to say anything about that. We in fact, we're telling you that we're not saying anything about that because we're purely.

300 "" (0) 01:46:46.430 --> 01:47:06.430 Defining voluntary preferences and there will be a range of things that will influence how you decide to, to respond to them. And I think there are avenues for expressions of, lists like these where groups can, you know, can draft them and put them out.

301 "" (0) 01:47:06.430 --> 01:47:15.530 Can comment on these specifications from the perspective of the stakeholders that they represent, whether they're cultural heritage, institutions.

302 "" (0) 01:47:15.530 --> 01:47:32.269 Or or or or or people that are addressing you know accessibility issues. I, I think that that those could also be more authoritative sources who can consider like what they think, a reasonable approach for their community might be. I.

303 "" (0) 01:47:32.269 --> 01:47:49.429 Don't know that we have the mandates to to speak for them either in this room, certainly not, not in a very broad way. Thanks Brad Glenn. So, I don't think we should have this section at all.

304 "" (0) 01:47:49.429 --> 01:48:09.429 I think that it crosses the line where we are trying to speak way beyond the reach of the adoption, the implementation of the preferences we are creating, and it violates our charter. Our charter is to create a our charters create a vocabulary.

305 "" (0) 01:48:09.429 --> 01:48:25.579 And a, a mechanism for attaching and communicating. It is not for starting to set policy about internet access to content or not access to content or anything like that or even communicating.

306 "" (0) 01:48:25.579 --> 01:48:41.119 This group's vision of how the internet should evolve in terms of access is merely AI preferences and how to communicate them. I, I will point out that, you know, I am authorized by my company to be here to work on an engineering technical specification of how to do those two things.

307 "" (0) 01:48:41.119 --> 01:49:01.119 I am not authorized to negotiate on, you know, the terms of, a sort of a tantumout license that would apply if we went with robots proposal, which essentially says, well, you know, AI preferences lets you do this, but we are carging out this exception here at the ITF that lets you do all of these.

308 "" (0) 01:49:01.119 --> 01:49:21.119 Other things for the good of the internet. I may like those things, by the way, but I'm not authorized as an engineer it's coming here and talk to you from my industry on whether those exceptions are blank exceptions we can grant. We shouldn't have this in the spec. It it I think it undermines the spec and frankly, if we were to produce a spec that has too many robust proposal in it.

309 "" (0) 01:49:21.119 --> 01:49:34.909 I propose or suspect that a lot of companies would simply say, oh, NO, I will not use AI preferences because it's bringing those requirements in with it. And so I'm gonna walk away and, and actually band the use of it.

310 "" (0) 01:49:34.909 --> 01:49:54.909 Thank you Glenn, Justin. I think that if you look historically, the way this has been argued in a legal sense, it didn't matter what was said because everything's arguable at the end of the day. So I think that if we try to carve out something that sets what's allowed and what's not, you know, you're not we're not supposed to do that.

311 "" (0) 01:49:54.909 --> 01:50:14.629 It really doesn't matter what we put in there because the reality is if somebody wants to argue it, they're gonna look at it. So I think what we're trying to do is carve out the spirit of the rules when it comes to respecting the preferences, but we're never gonna resolve the legal issue. If somebody wants to ultimately litigate it, they're going to litigate it as they have when it comes to robots and it has gone both ways.

312 "" (0) 01:50:14.629 --> 01:50:30.799 Thank you. Thanks. Yes. Brad, you, you spoke about liability and I was wondering if you could speak more specifically about where.

313 "" (0) 01:50:30.799 --> 01:50:48.409 Kind of where the increase in liability is with my proposal, the original 3.2 section list is 3.2 section versus kind of the rest of the document. Yeah. I think I'm aware of a couple of instances where.

314 "" (0) 01:50:48.409 --> 01:51:08.409 Certain countries have decided to create exceptions which cover all a part of these and others have not. And I don't think that there's a sort of consistent global norm around, e.g., deciding whether or not you can crawl and scrape content for accessibility purposes, for cultural cultural.

315 "" (0) 01:51:08.409 --> 01:51:15.439 Go to the institutions, I mean the US you know libraries got together to try and reform section one oh eight.

316 "" (0) 01:51:15.439 --> 01:51:35.439 To see if there were more, you know, exceptions or exceptions could be amended to, to update for, you know, or or for digital uses and their group didn't come to consensus. Fair use has adapted and adopted itself in, you know, many ways to cover certain of those, which is probably a reason why the group couldn't come to consensus, but then again.

317 "" (0) 01:51:35.439 --> 01:51:38.569 It kind of leaves the contours of where.

318 "" (0) 01:51:38.569 --> 01:51:58.569 The law decides certain things are or are not permissible, unclear, and I think that these, these are, are, are framed quite quite definitively. I know we use words like mights, but there's also, you know, there's also a pretty definitive you know reference to, to.

319 "" (0) 01:51:58.569 --> 01:52:04.789 To these not applying in, in, in, in certain circumstances. So I.

320 "" (0) 01:52:04.789 --> 01:52:22.369 I, I just think that someone shouldn't, shouldn't, shouldn't rely on these without a sort of more foresome set of advice from, for those who kind of looked at the relevant areas of laws that may apply and decide it's ok to do this or not. You know, this is just.

321 "" (0) 01:52:22.369 --> 01:52:40.969 Very, very much apart from whether or not you should be norms. I just don't think we can make them the norms. Thank you. Hi, I just wanted to echo one thing I saw in the chat that I thought was really helpful, and I think that.

322 "" (0) 01:52:40.969 --> 01:53:00.679 Including this list or some version of this list of exceptions that for very good reasons have been historically honored is really important to preventing a new round of fights over weather example or whether e.g..

323 "" (0) 01:53:00.679 --> 01:53:20.679 You know, accessibility uses might trigger some kind of violation or whether it somehow takes you out of someone quote unquote compliance with the preferences as they've been expressed. And I think that's important for two reasons. One is in some cases this would stand alone stand alone from the technical copyright regimes and what I'm.

324 "" (0) 01:53:20.679 --> 01:53:40.219 Or jurisdiction you're operating in, but also I do think there's a real and a legitimate role for us as the group drafting these preferences to make it clear how we think they do and do not apply, right? And if we do think that this is an example.

325 "" (0) 01:53:40.219 --> 01:54:00.219 As they've been recognized to be in, I think that we're in agreement are probably good instances where they may not keep following for good reasons, right? With out of technical compliance with the entire standard, I think that's helpful to clarify for people. Otherwise, there's a risk that you overchill activities that I don't think anyone here knows.

326 "" (0) 01:54:00.219 --> 01:54:18.409 Necessarily or maybe they do, but from what I'm hearing it doesn't sound like as a group we're necessarily looking to straight up band or discourage across the board. Well? Yeah I think.

327 "" (0) 01:54:18.409 --> 01:54:34.159 We, it's gonna be really hard to find consensus here and listening to those two discussions, the one we had before the group break and this one, they are essentially mirrors of each other.

328 "" (0) 01:54:34.159 --> 01:54:49.279 And I, at risk of sounding like a broken record, like I think the language that we have in here, the one about not prejudicing applicable laws, copyright exceptions and limitations.

329 "" (0) 01:54:49.279 --> 01:55:06.319 Is probably the lightest touch that we can introduce in saying these discussions, these balances are struck in other forum and we defer to those processes. I also want to point out here that.

330 "" (0) 01:55:06.319 --> 01:55:26.319 Even though I sort of like on substance in my personal views, fully agree with the language that's been proposed by either Terminalyssa, like even this historical historically in the in the language of Alyssa is very specific to.

331 "" (0) 01:55:26.319 --> 01:55:43.879 Some jurisdictions that are well represented here and probably isn't even true for a lot of others. Like we we are dealing with something that applies on the entire internet. I don't think we are in a position to even.

332 "" (0) 01:55:43.879 --> 01:56:00.049 Express something that is true Unless we really radically say we defer to other four to strike these balances. And do you think there's some text that goes along with your thinking?

333 "" (0) 01:56:00.049 --> 01:56:20.049 That's already in there that's or that's what like I, I think in Brad's sort of like thing it's slightly modified, but that is essentially and it's sort of very dense in legal language, but that is the the expressing preferences is without prejudice to and does not override applicable austin agreements.

334 "" (0) 01:56:20.049 --> 01:56:23.569 Including the applicability of exceptions and limitations to copyrap.

335 "" (0) 01:56:23.569 --> 01:56:43.569 That is the one sentence saying like we take guidance on these things from the legal arrangements that exist outside of this. Yeah, we can. Thank you. Kevin? Yeah, you know, I I worry too that we're not going to reach consensus if we need to delineate the.

336 "" (0) 01:56:43.569 --> 01:56:55.849 Exceptions or delineate every use that a publisher might want to prohibit. And as I said I I'm a fan of, of what Alyssa's attempting to do. Perhaps one way of.

337 "" (0) 01:56:55.849 --> 01:57:13.609 Again, I I believe Alyssa is simply stating a historic fact that is important context how to read this, but it does favor only the pro social positive reasons to ignore. Perhaps we could come to agreement on something that said.

338 "" (0) 01:57:13.609 --> 01:57:29.509 Historically, these preferences have sometimes been ignored both for these reasons, you know, that are, that are benign or or beneficial to the internet, as well as by bad actors for bad things.

339 "" (0) 01:57:29.509 --> 01:57:49.509 We are, this is this is gonna be NO different. We are not expressing. These aren't it's just it's another way of making clear that these are preferences and NO one should treat them as something that is more than a preference. Just to clarify, are you suggesting that we have a list labeled good and a list labeled bag? No, good. Thank you. Thank you.

340 "" (0) 01:57:49.509 --> 01:57:51.469 Thank you. Thank you.

341 "" (0) 01:57:51.469 --> 01:58:11.469 I mean I think we should say, you know, both good actors and bad actors for a variety of reasons including blah, blah, blah as a mixed list. Yeah, thank you. Martin, did you want to say? No, I, I like this framing. Okay, thanks. Keleb, you had a hand up and a couple of times and pulled off. Did you wanna say something safe? I mean I feel like Tori in particular addressed part of what I was going to.

342 "" (0) 01:58:11.469 --> 01:58:28.429 Say which is just to make a more narrow point, I don't think it's outside of the scope of this specification to delineate the context in which it's operating. In case that context changes later and just save us time.

343 "" (0) 01:58:28.429 --> 01:58:47.359 Litigating stuff in the general informal sense that as a society that we're gonna come to a conclusion on anyway. So, you know, that that would be my reason to support one or the other of these.

344 "" (0) 01:58:47.359 --> 01:59:03.979 In versions of this is just to sort of save us all time and establish a baseline that, you know, accessibility is important and we're operating we're we're making this the standard in that context isn't isn't it, is I think a valid thing to point out about the standard when we make it.

345 "" (0) 01:59:03.979 --> 01:59:21.109 So is there an expression that we've talked about that, that best gets there in your mind? Ooh, I don't know. In terms of the different proposals we've been talking about. I don't have a favorite. But it sounds like you're supporting saying something. I think so. Okay, yeah. Nate?

346 "" (0) 01:59:21.109 --> 01:59:36.349 I, I think Paul's points are persuasive, I think. You know, it's anytime you're baking norms in even ones that we think are social good, you still are making a judgment and it's I'm really like like.

347 "" (0) 01:59:36.349 --> 01:59:56.349 Accessibility is one where it's very obviously that something that we would probably most people would say is good but when you talk about malicious content detection, as I mentioned yesterday, you know, for some people, like, that can be received as sensorship, and so it's And this happens a lot actually on the internet, where people are are.

348 "" (0) 01:59:56.349 --> 02:00:02.989 Using sort of spam or malicious content detection and the response is NO you're sensoring me, and so we're weighing in on that.

349 "" (0) 02:00:02.989 --> 02:00:21.679 And I think that just kind of with all norms that gets difficult. I do know the language that's in there has a specific call out for exceptions to copyright, and I don't even think we need that. I think it can be as simple as just saying, the preferences the you know this is intended as an expression of preferences of the declament.

350 "" (0) 02:00:21.679 --> 02:00:37.789 And that's it. I mean that's all it is. It's, it's an expression of the preferences of the person who declared them, and that's, that's all we're chartered to do and I think that's all. So I just I just want to know that like it's only a 2 min robot strap that has this malicious content thing, right? The other.

351 "" (0) 02:00:37.789 --> 02:00:55.369 All the other three options don't have it. Like, so I'm not specifically opposed to that. I was just illustrating that like when you're baking norms in, even probes, probes sort of social norms, it's actually you're they're still norms and they're two sides of that. We're gonna try and wind the discussion up pretty soon, so.

352 "" (0) 02:00:55.369 --> 02:01:15.369 Yeah. Can I just react to this 1 s? Yeah, sort of, you started by saying like it was persuasive. I don't think like I find myself in your summary that it's, that you can remove the reference to exceptions and limitations there. Yes, copyright is rights of the current party.

353 "" (0) 02:01:15.369 --> 02:01:18.559 It's also rights of users to do things like.

354 "" (0) 02:01:18.559 --> 02:01:38.559 If if if you want to focus just on sort of like the rights side of shoulders, like then this is misrepresenting what I suggested for me like the exception limited exception limitation is super important in there. Okay, I I guess my, my point is is that it just seems why why that.

355 "" (0) 02:01:38.559 --> 02:01:40.759 One and not other norms.

356 "" (0) 02:01:40.759 --> 02:02:00.759 Yeah because we're referencing copyright on sort of our copyright is essentially a system that tries to strike balance in just pointing to one side of that balance that does not fully reference that. That's pretty standard language in, in many references to to copyright from other systems, so.

357 "" (0) 02:02:00.759 --> 02:02:05.059 Do do we Where do we reference copyright?

358 "" (0) 02:02:05.059 --> 02:02:20.989 Only in that very 1st section, in that text that we're referring to, that maps had proposed to change. Okay, I guess what I would say to, to clarify then is I would just build on that and say it should be even simpler and just be preferences like.

359 "" (0) 02:02:20.989 --> 02:02:35.989 If if we're not doing norms, we shouldn't do any norms. Like. I think Nate it'd be better if you can suggest like changes like either as an issue or like something like that. Like I think it's it's easier to discuss things when we know exactly like what changes you wanna make Glenn?

360 "" (0) 02:02:35.989 --> 02:02:55.989 So I'm in favor of removing it entirely. Moving the mentions the section and the mentions. I I could live with what Nate is proposing of radically reduced things abnorms. I think we are very much at risk of exceeding our mandate here in talking about this topic area.

361 "" (0) 02:02:55.989 --> 02:03:16.119 And if we were to go down this path, we have a very pro open internet voice in this room, the ITF is a very pro open internet voice as well. But if we wanted to be fair, we would also have to start inviting in other parties with other views to comment on this if we wanted it on that path, I think it's a very dangerous path because we, we straight into policy.

362 "" (0) 02:03:16.119 --> 02:03:16.789 Space.

363 "" (0) 02:03:16.789 --> 02:03:36.789 We are very much gated out of the role of what preferences can be expressed, ok, and how do you express them? That was our mission. This, all of this other stuff is out of our scope and and to draw an analogy, if we're sitting here talking about a DNS specification, and somebody said, I want to put a sp.

364 "" (0) 02:03:36.789 --> 02:03:41.359 Rule this DNS, I won't put this is just commentary, don't worry about it. It's just commentary.

365 "" (0) 02:03:41.359 --> 02:03:57.979 And the words I want to put in are that the DNS should be really open, it should be free, but it shouldn't contain any hate speech. So any domain name has hate speech shouldn't be allowed. Okay, that's great. That's a very simple statement, and I think it's one of the statements that a lot of people go, Yeah, hate speech is bad.

366 "" (0) 02:03:57.979 --> 02:04:14.239 But it carefully avoids the definition of what is hate speech, right? It opens up a lot of cans of worms that are outside the scope of a DNS specification. Just like this opens up a lot of cans of worms that are outside the scope of an ITF specification, how to express a preference.

367 "" (0) 02:04:14.239 --> 02:04:34.239 If ultimately people in this room want to later on after the specification is published, how you express the preference, want to write articles, proses, blog posts, feces that get published on what they meant and what it interp the interpretation should mean, albeit, have fun, but it shouldn't be in an RFC specification within.

368 "" (0) 02:04:35.029 --> 02:04:55.029 So I think we've had a lot of discussion here about the substance of this, but I think concretely the direction that we've been heading in here is, to clarify exactly the point that a number of people have been making, we are justifying preferences and without.

369 "" (0) 02:04:55.029 --> 02:05:12.649 This section or something like it, at least in its minimal form. Now there's a bunch of options that we have in in terms of how much we, we intend to pave the path toward the things that we think are pro social or what have you. But without anything here.

370 "" (0) 02:05:12.649 --> 02:05:32.329 We failed to address the question that we continuously get about, is this a preference? And we need to retain some amount of text in this document that makes it clear to people that we are defining preferences and not rules. And if we were to remove the section entirely.

371 "" (0) 02:05:32.329 --> 02:05:48.499 We would, we would fail on that. And the reason that it was added is we've got continual confusion from people about this being preferences. And so that is the primary reason the section exists. Now, I think the the list of things that.

372 "" (0) 02:05:48.499 --> 02:06:07.969 Was was added in there that specific examples, I do think is a case of us putting a thumb on the scale to some extent, and I don't think this is a problem in the way that Glenn suggests. I think this is potentially ok if we find those examples to be acceptable.

373 "" (0) 02:06:07.969 --> 02:06:24.439 I'm tending to lean to, toward what, what Alyssa is saying as a as a way to get a balance out of, out of all of this, but there are a number of options for this and I could all all are possible. If anyone wants to look at.

374 "" (0) 02:06:24.439 --> 02:06:39.589 Some of the pull requests are up there, there's some rather options that I put up that take a lighter touch toward this thing, but I don't think that that really changed changes the substance. I think we either make a modification along the lines of what Alyssa has suggested.

375 "" (0) 02:06:39.589 --> 02:06:55.279 Or we do that plus having the list of items, I don't think we can get away with nothing. Okay, thanks and did you have any thoughts on Brad's stuff?

376 "" (0) 02:06:55.279 --> 02:07:15.279 So I think there are some useful statements in in Brad's thing in terms of further clarifying the the very short statement that we have in the introduction, and one of the things I would like to see is that being used in addition to the other things but not as a complete replacement for the section. Cause what I think Brad is addressing is.

377 "" (0) 02:07:15.279 --> 02:07:34.579 Is different, a different aspect of this problem, which is to say that text that you suggested is an expansion of the text in the introduction that is a lot clearer about the, the, the intent of purpose. And I think that's a useful addition, but I don't think it addresses the core.

378 "" (0) 02:07:34.579 --> 02:07:49.939 Question of whether someone receiving the preferences isn't, is, is forced to be in a position whether they choose to honor them or not. And we have to be able to say that in order to address the questions that we received thus far and.

379 "" (0) 02:07:49.939 --> 02:08:07.159 You're talking much more about the legal effect that that does or does not have. And I think it was useful clarification Paul and I discussed that and there's some really useful text there that we could add, but I don't think it's a replace. Is the drop off?

380 "" (0) 02:08:07.159 --> 02:08:27.159 And then we're gonna wrap this one up, I think. I think I'll pass, thank you. Yeah, give us just 1 min, come into my office.

381 "" (0) 02:09:07.159 --> 02:09:09.649 So.

382 "" (0) 02:09:09.649 --> 02:09:29.649 This text was was added to the specification in a simpler time, a suggestion was made to the editors and they they added it to to to see discussion. It's pretty clear we don't have consensus on the text yet, so in that sense this text is at risk in the specification.

383 "" (0) 02:09:29.649 --> 02:09:49.609 I'm hearing, the most agreement that something needs to be said about the relationship of of preferences to the legal context, that very short bit at the top of what we currently have. The rest of it is I think still in play, and I think we need to continue those discussions.

384 "" (0) 02:09:49.609 --> 02:10:09.609 So, we'll, we'll continue on the list, I think, but underlying all of this, I think there's attention that on the one hand, we want to make sure people are aware that preferences are not instructions or you know rules that need to be followed.

385 "" (0) 02:10:09.609 --> 02:10:14.689 For a control mechanism. On the other hand, I think, you know.

386 "" (0) 02:10:14.689 --> 02:10:34.689 There's a concern if I could summarize that a language like this in the specification gives people an excuse to say, well, NO, I'm following the specification and it says I can do that and therefore I don't have to, you know, follow the the preferences. And especially when we have certain jurisdictions going and saying that part of people's legal obligations are to follow.

387 "" (0) 02:10:34.689 --> 02:10:54.689 Preferences this kind of, you know, over eggs the pudding and and it makes a decision in this body that maybe is more appropriate to be making somewhere else. So I think we have to, to, to talk about solutions that are cognizant of both of those motivations and find a way to weave our way through the concerns on both sides. So probably a lot more discussion.

388 "" (0) 02:10:54.689 --> 02:11:14.299 I would be surprised if the text in its current form survives, but let's let's see how we go. Yeah, and I think Echo brought up a, interesting point in the chat, like which is not the denying or something, but actually permission of something. He wants to be able to interpret it without like.

389 "" (0) 02:11:14.299 --> 02:11:32.059 Deviating from it. So Echo, did you want to kind of talk through it? Because I think I have found it very interesting that you took the other side of this discussion. Sure, I mean, I mean like the point of this exercise from my perspective is that when the application goes and examines this, this examines this, you know.

390 "" (0) 02:11:32.059 --> 02:11:52.059 The whatever thing that's attached in terms of preferences on the site, then, you know, they're, they're entitled to do something with it and they and like they can they can they have some knowledge with their what what people want them to do or not want them to do. And like.

391 "" (0) 02:11:52.059 --> 02:12:12.279 So imagine that imagine there's some jurisdiction whose background rule is that you can't do AI training without permission. If I go to the site and it says foundation model equal equals yes, then I need to be able to interpret that as actually that has permission to train. And if that's not true, then I understand what the point of any of this work is. And so the text that's on the screen right now.

392 "" (0) 02:12:12.279 --> 02:12:14.629 That like says basically like.

393 "" (0) 02:12:14.629 --> 02:12:34.629 Without like, you know, blah blah blah expressing presses out prejudice too it doesn't override blah, blah blah. Like, I don't say that could work because it seems to contradict what I just said. So, now people think that like it is the case that in that jurisdiction, like I'm not entitled to rely on financial model training equal equals yes I'd like to hear them say that because I don't then I don't understand what the proper.

394 "" (0) 02:12:34.629 --> 02:12:52.009 So, so this is a proposal from just to said that sure I agree. Yeah I agree. Gotcha thanks. All right, I have to go look at the hour test. I'm not sure if the air text has this problem. Thank you. Maggith, do you want to say what you said in the chat?

395 "" (0) 02:12:52.009 --> 02:13:09.259 Oh, well, I think there's just a circularity here because on, on one hand we're saying these are just technical preferences, they're about just technologically expressing and understanding the preferences, so we can't talk about norms or laws or stuff outside of this technical preference.

396 "" (0) 02:13:09.259 --> 02:13:25.489 And then on the flip side saying, we can't have talk about which reasons you might not because that will take our legal obligations. Like either these potentially change our legal obligations or they don't. And I think you you sort of have to accept that they are.

397 "" (0) 02:13:25.489 --> 02:13:45.489 Strictly speaking, preferences, but will also shape people's understanding of their legal obligations. And to try to argue that it's all one or all the others isn't realistic. So I I think the point is like there's different people saying these two arguments, right? Like, so it's not the same person's trying to state both of these things. I think that's kind of what like.

398 "" (0) 02:13:45.489 --> 02:13:55.969 Don't have consensus on it because some people want your former thing and some people want your ladder thing. It's kind of at least how I see it, that is fine.

399 "" (0) 02:13:55.969 --> 02:14:15.969 Okay. So we had three other issues here that I think are are probably will take less time. Oh, do you have more people on that? I think Brad put his hand up and Kevin did so if you want to talk talk. Oh, you're ok? Yeah, I mean I think it will just be the full back and forth about this and we've done it. Okay. Thanks, kevin.

400 "" (0) 02:14:15.969 --> 02:14:35.449 Yeah, just in the interest of making progress, it seemed like there was although we have not reached consensus, there was more consensus around an approach like Alyssa's than what is currently in the draft. Does it make sense to check that assumption and do a quick poll so we know where we're starting from? I was thinking of doing polls.

401 "" (0) 02:14:35.449 --> 02:14:55.449 I, I think I'm reluctant to pull it because we have a number of proposals out there right now and I I think a beauty contest isn't gonna help. I I would agree with the characterization that something more minimal is is probably where we're leaning, but it's, it's perhaps.

402 "" (0) 02:14:55.449 --> 02:15:13.039 In the discussion to make a definitive statement. David Robert. Yeah, I think there's a variety of things being discussed and I think one way to potentially make this a little bit easier would be to try to separate.

403 "" (0) 02:15:13.039 --> 02:15:28.159 The different goals of this section that different people have, right? Because I, like I suspect, e.g., trying to clarify that these are preferences is fairly low.

404 "" (0) 02:15:28.159 --> 02:15:43.759 Intention and then it's the, you know, the additional parts of that that people have different issues with or favor towards, and so if we chunk it up, we might be able to.

405 "" (0) 02:15:43.759 --> 02:16:02.959 You know say oh this this part is fine and we're actually only arguing about these other parts. Thanks for that. Yeah, I think I think So 1643 .1 conformance.

406 "" (0) 02:16:02.959 --> 02:16:20.329 Joe suggested removing this section, I think, and we talked about this in March. I think we're at a place where this is something that we just need to to improve the language sum. So I think we assigned this one to the editors.

407 "" (0) 02:16:20.329 --> 02:16:38.899 Actually have a Do you have a proposal? Yeah the support request after that if you want to bring that up, I forget the number, 1905 maybe. Is it linked? It is not that should be. There's not.

408 "" (0) 02:16:38.899 --> 02:16:55.939 1905? Try 1905. Yeah, 95. Yeah, look at that. Pretty good. Rain works. The network doesn't. You know, it doesn't.

409 "" (0) 02:16:55.939 --> 02:17:15.939 No, this is the network. There it goes. Okay, so there's there's a bunch in here but you should look at the words, not the right now because the words are easier. Where the diff? The diff. And it's basically a wholesale replacement for the section. The whole point of this is to say.

410 "" (0) 02:17:15.939 --> 02:17:18.499 The section 3.1 is addressing.

411 "" (0) 02:17:18.499 --> 02:17:36.649 About addressing the question of what the purpose of the specifications primarily is, which is making sure that when someone expresses a preference, there's a good chance that the person receiving their preference understands what their preference is. It's not about whether they choose to conform or not. That's section 3.2.

412 "" (0) 02:17:36.649 --> 02:17:53.779 Section 3.1 is addressing the question of, can someone expressing their preference be confident that the person perceiving it understands what has been said? There are the three problems with that. 1st one is the vocabulary definitions that we continue to debate.

413 "" (0) 02:17:53.779 --> 02:18:10.759 I've not addressed that in here. Hopefully we can, we can get to a point where the vocabulary is clear enough that the people at least agree on what the meaning of the, the, the categories of use are. What I have put in here is a couple of points about the limitations.

414 "" (0) 02:18:10.759 --> 02:18:28.009 Aside from that. And one of the limitations is that a recipient will only understand preferences expressed through mechanisms that it has imple implemented, and, that might be the stuff that we define in our, in our attachment document or it could be other methods.

415 "" (0) 02:18:28.009 --> 02:18:44.809 And so if a preference preferences are associated with assets that using other means, then they simply won't be understood. That's just a statement simple statement of fact. The other point that is worth raising here is that the.

416 "" (0) 02:18:44.809 --> 02:19:03.019 The attachment mechanisms that we referenced there don't always do things to authenticate where the preferences come from. And so you can't make any assumptions about that unless there's something explicit in the attachment mechanism to say this preference comes from this entity.

417 "" (0) 02:19:03.019 --> 02:19:23.019 And that's an that's something that would be essential in a registry based system. If you're operating a registry based system, you have to build in things like authentication, but the methods that we're defining basically are only implicit. So e.g., preferences in robots.text only imply that the operator of the server is the.

418 "" (0) 02:19:23.019 --> 02:19:32.689 Most of the preferences, but they could be doing some on behalf of expressing those preferences on behalf of others like the original creator of the content or or other things like that.

419 "" (0) 02:19:32.689 --> 02:19:48.919 And so that's what the last two two paragraphs here say. So it's a little longer than it was before, but I think it's, it's more precise. And so my my hope is that this is clearer than what we currently have and, and sort of.

420 "" (0) 02:19:48.919 --> 02:20:08.919 Outlines the sort of philosophy behind the the work that we're doing. Thank you. Yeah, so I mean, intuitively just at 1st glance, this looks like an improvement. I guess my initial inclination would be to let people have a look at it, take it to the list.

421 "" (0) 02:20:08.919 --> 02:20:21.709 Get some feedback and then, do a, a a little call for consensus to incorporate it because it is a fairly substantial bit of text. But if folks have initial reactions or thoughts in the room, we, we can, we can hear that.

422 "" (0) 02:20:21.709 --> 02:20:41.709 People probably haven't had a chance to look at it deeply. It on 1st glance it looks good, so thank you. Anyone want to say anything at this point or should we just take it to the list? This sounds like a good idea. Give us a chance to digest it. Okay, I'll send an email to the list.

423 "" (0) 02:20:41.709 --> 02:20:43.759 Thank you, that's great. And just.

424 "" (0) 02:20:43.759 --> 02:21:03.759 Just yeah, make it clear that we'll have a call for consensus. Can I ask one question of the who you're in the room? Sorry I can't hear you. I'd like to ask one question while we're all here in the room. So, question on this. By the way, I'm not opposed to this text. What do you anticipate the.

425 "" (0) 02:21:03.759 --> 02:21:05.900 The takeaway or the action.

426 "" (0) 02:21:05.900 --> 02:21:25.900 Of the reader of this text to be. I don't disagree with what you've written in here, but I'm then asking myself, ok, what does the follow on action anticipated of the, the, the reader? Yeah. So the point is to understand what it is that they're getting from this document when when you implement it, right?

427 "" (0) 02:21:25.900 --> 02:21:35.030 And, one of the problems that we've had with a lot of these things is that people don't understand the limitations of the things that we're defining.

428 "" (0) 02:21:35.030 --> 02:21:50.390 And, specifically here, this, this is addressing the question of what is the purpose of this document, and the purpose of the document is to make sure that someone expressing a preference can be understood. And that's all it says. But it's putting limitations on that.

429 "" (0) 02:21:50.390 --> 02:22:10.390 By saying, if you don't understand the vocabulary terms, if you don't understand the attachment mechanism, then you're not gonna be able to understand the preference. That's just a, it's a simple statement of fact, Nothing more than that. I I wonder then if, if even the word limits might limitations to be aware of something to be added to one That's.

430 "" (0) 02:22:10.390 --> 02:22:30.390 Okay, take a look and if you have suggestions on the PR or yeah. I mean in in the past especially when we 1st started, we talked you'll probably remember we talked about maybe having an applicability statement to guides interested policymakers if they want to use it, their regime, that seems to have fallen by the wayside, but this.

431 "" (0) 02:22:30.390 --> 02:22:34.940 Kind of is maybe trying to fulfill that need a little bit, the sound of it.

432 "" (0) 02:22:34.940 --> 02:22:52.310 It is also sort of the previous language was much more prone to misunderstanding, like calling it conformance sort of that I think a number of people trapped up and thinking that was actually about compliance. Right. So.

433 "" (0) 02:22:52.310 --> 02:23:10.430 It's also like a simple like like look at the changes to the previous section. It made sense to a policy sorry to a protocol designer, but probably not. No, it didn't even make sense to me when I looked at it 2nd time. It was just NO good. All right, so I think I think that the two bullet points here are the ones that.

434 "" (0) 02:23:10.430 --> 02:23:29.150 They're about to really crisply summarize what's going on. So Martin I link to the PR from the issue, ok? Just in case anybody follows there. I was gonna do the same, but fantastic. Okay. And we also had, 1607 distinguishing access and use.

435 "" (0) 02:23:34.580 --> 02:23:51.590 So this is another issue where I think the resolution is mostly editing work, looking through the issue now.

436 "" (0) 02:23:51.590 --> 02:24:09.440 Did did the editors make any progress on this? Yes. But I think it's worth having it having a discussion here. Does anyone have a problem with the basic understanding?

437 "" (0) 02:24:09.440 --> 02:24:26.600 Yeah, but there's there's two mechanisms to play. One is if you're a crawler or a or or a scraper or whatever it is, you obtain some content. That's, that's the the mechanism that we're not talking about.

438 "" (0) 02:24:26.600 --> 02:24:42.680 The thing that we're talking about is once you've obtained something, how you then subsequently use that thing? Is that something that anyone finds challenging or would disagree with? Sorry, can you repeat?

439 "" (0) 02:24:42.680 --> 02:24:58.250 So there's, there's, there's two things that someone might do with a piece of material. Obtain it is one thing and then use it is 2nd thing. We're separating those two things out. And robot so text governs.

440 "" (0) 02:24:58.250 --> 02:25:14.720 The acquisition of content, assets, material, whatever, what we're doing is then downstream use of that and expressing preferences about the use, not about the acquisition.

441 "" (0) 02:25:14.720 --> 02:25:34.720 Robuster text specifies preferences about the acquisition. We're not changing any of that. We're just addressing the use downstream of that. Does anyone have a problem with that framing? Because I think otherwise this is just an an editorial exercise on our part to make sure that it's very, very.

442 "" (0) 02:25:34.720 --> 02:25:53.960 Very clear. Clear communication, yeah. And I'm I'm actually struggling a little bit because I think it's it's clear enough to me from reading it and I did try to address this. So maybe it's just other people including the chairs, need to go and have a look and make sure that they agree with you that it's clear now.

443 "" (0) 02:25:53.960 --> 02:26:09.710 Right. I think Ronnie Ronnie, did you want to Yeah just regarding this clarification, so I agree with the differentiate the two parts and not necessarily any wording that will.

444 "" (0) 02:26:09.710 --> 02:26:29.710 Say something about what governs the, the access part. So just calling out saying we are not talking about accessing what we are talking about the usage. That should be it without saying what does govern it or or any other aspects of that. Okay, Thanks yeah NO just to let everyone else know but it's it's.

445 "" (0) 02:26:29.710 --> 02:26:49.710 Flagged at the bottom of this issue. There's a related issue, which is the other one you had on your list, which is the box collect data for multiple purposes that I believe resolving this will also resolve that one and we can get a two for one, right? In this case. And and so I think that's the next on our list to discuss. I believe it is, that's right? Yeah.

446 "" (0) 02:26:49.710 --> 02:27:09.730 So yeah, is there anything else to talk to talk about here? Hecker said he was gonna take a look at this one. Yeah, I ad quickly failed to take a look at it, so I think we can just we can close this issue and if I book at it later on.

447 "" (0) 02:27:09.730 --> 02:27:11.180 I hate it, I'll reopen it.

448 "" (0) 02:27:11.180 --> 02:27:31.180 Okay, I'll close it now then. Love closing issues. Let's do more of that. Okay, it's noon. We've gone through that. What I suggest is we go and take a a lunch that is shorter than yesterdays, come back and we can have a time box discussion.

449 "" (0) 02:27:31.180 --> 02:27:49.250 About the use issues, I'm gonna time box it because I want to get back to what we discussed yesterday, which was the the training issue and the search issue and the proposals we have there. I think that's a very productive discussion. For the use issues.

450 "" (0) 02:27:49.250 --> 02:28:09.250 We we have, I think Bradley's proposal, but that was made a long time ago now, and there's been a lot of water under the bridge and a lot of discussions around there. So what I think the purpose of a used discussion would be to talk about, you know, the concerns, the objections, the context of, of, of what we're thinking around.

451 "" (0) 02:28:09.250 --> 02:28:11.750 On that so that people can go away and make.

452 "" (0) 02:28:11.750 --> 02:28:31.750 Proposals that are more likely to be successful in our next meeting, and and and yes, that's foreshadowing. And when we talk about use I think a lot of you know people are also want to return to the discussion of a top level term, a lot of the concerns around use also apply to that top level term, so maybe we can get a bit of a two for there and.

453 "" (0) 02:28:31.750 --> 02:28:43.550 And give people some thoughter for, for, for proposals in those areas. But we'll time box it because we don't have substantive proposals that are current that we can talk about right now, and then we'll return to to the.

454 "" (0) 02:28:43.550 --> 02:29:03.550 To to the training and and search issues. Does that make sense? So let's let's just take ah honest 1 h lunch. In other words, be here and actually starting discussion again at 01:00 p.m.. So please re enter the room five or 10 min before that and settle yourselves.

455 "" (0) 02:29:03.550 --> 02:29:21.890 Did you say to be back? Sorry? What times did you say to be back up so 01:00 P.M. 01:00 p.m.. M. 01:00 p.m.. M. Yes. That are in it.

456 "" (0) 02:29:21.890 --> 02:29:41.660 Oh, really?

457 "" (0) 02:34:36.380 --> 02:34:56.380 Oh Echo. Echo. No.

458 "" (0) 03:30:22.350 --> 03:30:42.350 Yeah. So.

459 "" (0) 03:30:42.350 --> 03:30:53.619 Okay, the meeting mics are back on, so whatever secrets you have?

460 "" (0) 03:30:53.619 --> 03:31:12.309 Okay, let's get started again. Krishna? We're getting started. John, here we go. Okay, so hopefully everyone had a nutritious and exciting lunch. Exciting. Well, I think.

461 "" (0) 03:31:12.309 --> 03:31:28.269 So we've got this afternoon and we've got tomorrow morning. We do intend to finish on time tomorrow. If not beforehand, we'll see how much energy you folks have. As I said.

462 "" (0) 03:31:28.269 --> 03:31:46.749 The only issues we haven't touched so far are these last 2172 and 01:50 about a use category. And so what we thought we'd do is have a fine back discussion about those with the the main purpose being to, to, to refresh everyone's memory.

463 "" (0) 03:31:46.749 --> 03:32:04.539 And to, give folks who want to make proposals in this area or refresh their proposals in this area, some some information that might help them make those proposals more likely to gain consensus. So let's spend 1 h on this so we'll, we'll, we'll take up until around two.

464 "" (0) 03:32:04.539 --> 03:32:24.539 On this discussion or or up to that. If we end naturally before that, that's fine. And then we'll go back to the discussion of the training term and the search term. So 172I think is is fairly straightforward. It is not a proposal, it's just pointing.

465 "" (0) 03:32:24.539 --> 03:32:29.469 You have the need, oh NO it's not wanting to do anything. There it goes.

466 "" (0) 03:32:29.469 --> 03:32:48.399 The network is being slow. There's a new network? Yes, so from Chris pointing out with and this has been said a few times in a few different ways, without a category that covers rag and branding usage, the vocabulary does not meet publisher needs and that was talked about quite a few times.

467 "" (0) 03:32:48.399 --> 03:33:08.049 Right. And then we also have one 70, sorry, 01:50, which was, was open before that from Brad, which was a proposal for, a category to do that. Now, we we've had a lot of discussions since then.

468 "" (0) 03:33:08.049 --> 03:33:27.249 And and I think that it's clear that in its current form, this proposal isn't is likely to gain consensus, but we, we do want to kind of take a look at this and and look at the other discussions in this area and see if we can give these folks some input to help them shape what they're doing. So this was Brad's proposal I think starting right here.

469 "" (0) 03:33:27.249 --> 03:33:45.189 Brad, do you want to say anything to this now to kind of give us some context? Sure. So I think, you know, while a lot has changed since we since we last discussed this, the one thing that hasn't changed, which is, you know, certainly just.

470 "" (0) 03:33:45.189 --> 03:34:05.189 M, you know, more of what it is than it was back then is is how important this category is for, for publishers. It is certainly just in terms of volume, the crawling that is relating related to rag as opposed to crawling that is traced back to.

471 "" (0) 03:34:05.189 --> 03:34:11.739 To training, is, is more substantial, of the calling that.

472 "" (0) 03:34:11.739 --> 03:34:31.739 That is happening out there that crawling publisher sizes what's is what's driving the majority of crawling, increasing crawling, and rag is, is the single biggest influence of what's driving what's driving the proportion of crawling.

473 "" (0) 03:34:31.739 --> 03:34:36.219 Between training and and and post and post training.

474 "" (0) 03:34:36.219 --> 03:34:55.119 For us it kind of gets to the heart of how our content is kind of being scraped for the purposes of creating substitutive products and, linked directly to a drop in referral traffic. It seems that the referral traffic coming from.

475 "" (0) 03:34:55.119 --> 03:35:15.119 AI services that Crawlporag is actually getting, getting, getting worse, not better perhaps because the proportion of crawling is, is, is growing so quickly. Referral traffic is, is probably about .04 % of all referral traffic coming from.

476 "" (0) 03:35:15.119 --> 03:35:33.399 Coming from from from from AI related calling comes from, from AI chatbox and that's that's public information I can send around references to that. We can take a closer look at. So it, it seems that you know if we if we don't address this, we'd really be.

477 "" (0) 03:35:33.399 --> 03:35:50.829 Failing to address a huge proportion of what we're calling this happening, not just only the sort of qualitative concerns that that we've mentioned that had discussed a lot of times here before. And, and it it.

478 "" (0) 03:35:50.829 --> 03:36:10.829 Casting back to the previous discussions, if I could try and attempt to characterize, you know, some of the the the concerns that were raised about previous use proposals, maybe not about this specifically, but one of the issues we, we kept on wrapping ourselves around was that if if you have a really broad term like use that it it can apply to, you know, I think.

479 "" (0) 03:36:10.829 --> 03:36:30.829 You and a lot of other publishers have in mind specific things that that you you see as unfair or abusive or problematic, but there are other many many other uses of AI as a generic technology that may not be wrapped up in that and and and doing the line drawing exercise of of of segmenting those is, is very difficult. And if you just say.

480 "" (0) 03:36:30.829 --> 03:36:50.829 Use of AI in in one of, you know, if that's what our vocabulary ends up saying, that can capture lots of things like, you know, summarization tools on my own laptop and, and, and, and or or, you know, accessibility tools and and or or even just, you know, other, you know, maybe maybe they're not covered by specific exceptions, which we talked about before, but just generally local use.

481 "" (0) 03:36:50.829 --> 03:37:00.909 Uses versus you know maybe uses that can be characterized as republishing or whatever. And, and so coming up with language to try and draw that line is I think one of the challenges here.

482 "" (0) 03:37:00.909 --> 03:37:20.909 And and and for me, personally underlying all of this is this notion of, you know, AI is, is, is a technology, you know, it's it's not, you know, a particular purpose and we've talked about purpose based before. And so if you just say, don't use AI with my data, you're kind of saying something equivalent to don't use linked lists.

483 "" (0) 03:37:20.909 --> 03:37:30.219 Or don't use databases, you know, it's a computer science tool. It's not a purpose. And so if we can untie that, maybe that's a way we can.

484 "" (0) 03:37:30.219 --> 03:37:50.219 Makes some progress here. 1st time question. What was your 24 % referring? Zero four. That of this is this is this is a state from toll bits. You can look up their state of the bots report. They they estimate that AI.

485 "" (0) 03:37:50.219 --> 03:37:54.279 Applications only account for points zero 4 %.

486 "" (0) 03:37:54.279 --> 03:38:12.939 Of referral traffic for publishers, the bulk of referral traffic comes from Google. But and that's and that's a decline they've noted over their, you know, what they've calculated the last couple of quarters.

487 "" (0) 03:38:12.939 --> 03:38:30.699 Brad, if you can throw in the link for the thing if it's available publicly. Yeah. Yeah, so the other thing I wanted to which that this particular proposal covers in addition to what Brad mentioned is one that I.

488 "" (0) 03:38:30.699 --> 03:38:50.699 Many of you know have been very strongly in favor of or specifically my my company and and our users in favor of and that is the need for rights holders to be able to stipulate that for a given asset, it not be used for things such as stuff.

489 "" (0) 03:38:50.699 --> 03:38:56.139 Violistic reference or reuse.

490 "" (0) 03:38:56.139 --> 03:39:13.029 In that model, this is something that we actually provide today using C two PA and and capabilities, and it's implemented in our tooling and it's something we'd like to see continued expressible in the vocabulary from the IETF.

491 "" (0) 03:39:13.029 --> 03:39:33.029 And this particular wording enables that, as you can see in that 1st paragraph where it says including stylistic elements thereoff. But, but so that that the my immediate reaction to that is is does it make sense to bundle that up into this kind of generic use category? Is that something that's maybe more productive if we talk about it as a separate term? Yeah, I'm I'm not.

492 "" (0) 03:39:33.029 --> 03:39:52.269 Specific to how we do it. I just want to be I want to let for those who had not been part of previous discussions, make it clear that that is another one of the things expressed by this proposal. It's not just about rackles. It maybe interesting if you were to make a proposal that's for a separate term and, you know, see if that has more.

493 "" (0) 03:39:52.269 --> 03:40:12.269 We've I I seem to recall we tried that in previous meetings and we ended up saying that it was easier to bundle it, but I mean the group consensus to go back in that direction I'm not objectionable. I think to be my feeling would be to what you were saying previously, which is if we move from this generic use concept back to.

494 "" (0) 03:40:12.269 --> 03:40:27.249 Purpose or intent or wherever else we may end up going, then yes, I agree it may make finer grained sense, but I think I'll withhold that until we see where the group as a whole goes. Sure. Don't, don't wait for consensus to make a proposal though. No, NO NO NO, No I am with you.

495 "" (0) 03:40:27.249 --> 03:40:47.249 Nick? So I am one of the folks. This is a very important issue to me as I've previewed on the mailing list. I am very dedicated to it and I want to come up with a proposal. I'm here in Park to understand what the objections are and Marcus highlighted a few of those which I think is helpful. I've invited folks to reach out to me multiple.

496 "" (0) 03:40:47.249 --> 03:40:59.919 Couple times on the main list if they want to collaborate on this. I'd especially I I'd reiterate that and especially extend that to folks who share those concerns that mark or or other concerns about this category so that we could try to come up with something that.

497 "" (0) 03:40:59.919 --> 03:41:18.129 Gets us towards consensus. I do think that this is the critical future for the, the, the web and I, at least among the issues that we're tackling, and I actually think that if we don't do this, we're, we're actually not even fulfilling the charter because the charter specifically talks.

498 "" (0) 03:41:18.129 --> 03:41:36.189 About development, deployment and use. And right now we basically have a category, one category about development, and then we have the search category isn't even really a preference, it's like an, it's more of like an exception to that other category because we expect that everybody's gonna say yes to it. So we have.

499 "" (0) 03:41:36.189 --> 03:41:53.169 An incredibly limited amount of expressions if that's all that we do, and our charter specifically talks about use and the way that these models work, and I know there are machine learning people in here who are probably much smarter about this than I am, so feel free to correct me, but as I, my metric for understanding this is, basically.

500 "" (0) 03:41:53.169 --> 03:42:08.499 You know, these are really long math equations and you train up the foundational mount model as the base of that math equation, and then you put other tokens in a context window and then that the combined set of those.

501 "" (0) 03:42:08.499 --> 03:42:28.499 Tokens that math equation produces an output, and we have given away with the train category to express a preference as to everything up to foundation model production or whatever we're gonna call it now. But we have to do something about the other part of that equation. Like it seems, it just seems logical that we would need to address that. I understand.

502 "" (0) 03:42:28.499 --> 03:42:30.639 There are, there are challenges that there are.

503 "" (0) 03:42:30.639 --> 03:42:48.819 You know, many ways that this might apply, but that is the work that matters, and I would actually say that this is the one category we could, we could do something wrong that might actually matter. Like we don't need in a search category as an independent category, that would be the only category would be completely unnecessary. Search has been around for like 3040 years.

504 "" (0) 03:42:48.819 --> 03:43:08.819 Don't need that category. So really if we're only looking at the trained category, well the problem with the train category is by the time this actually has any effect, it's gonna be like 2028 2029 by the time we publish this, it goes through and the market adopts it and before new models are trained, we'll be seven to eight years into these, the training of these models and the way that they're trained is they're trained on prior.

505 "" (0) 03:43:08.819 --> 03:43:11.349 Versions of the models. So like.

506 "" (0) 03:43:11.349 --> 03:43:29.049 I I'm not saying it's not important and I I do think it matters, but to some extent training is a fate to comply and so if we're going to kick the can on context window usage down the road, we're gonna potentially end up with the same situation where we're only enabling expression the preference after those preferences like.

507 "" (0) 03:43:29.049 --> 03:43:49.049 Those preferences are useful, but even really useful in 2021 or 2022 to have a way to express a preference around trading. That usage that that the, you know, the importance of that has sort of or the, the utility that I guess has sort of declined over time. A couple other things I want to say real quick. This I think everything Brad said is absolutely.

508 "" (0) 03:43:49.049 --> 03:44:01.779 Absolutely true, and I agree with, I think that a lot of discussion has been about how this is critical for the sort of incentive to create content on the web. You know, I think if you care, you I think probably a lot of folks.

509 "" (0) 03:44:01.779 --> 03:44:21.779 Who aren't the publishing industry aren't always sympathetic to publishers and a lot of people don't like the media and I understand why, but if there's not an incentive, like, if the incentive is that you can only participate in, in, you know, that if the incentive structure becomes that you should put stuff behind walls, because that's the only way that you can control it or.

510 "" (0) 03:44:21.779 --> 03:44:42.239 Then that's going to lead to a more close web. I think you see that when you browse the web and you go to things and all the paywalls that pop up, and so we're gonna be incentivizing more of that. We should be incentivizing the publication of content openly and giving preferences to the creators who are doing that allows them the confidence of being able to publish openly, the effects that.

511 "" (0) 03:44:42.239 --> 03:44:50.709 What Brad's talking about or the data is really stark on how much lower the click through rates are from AI search summaries versus like traditional search.

512 "" (0) 03:44:50.709 --> 03:45:06.279 Chartbeak came out with a study yesterday that was published that said that for large media publications, they've seen on average a 22 % decline in search referral traffic just over the last two years. And that sounds bad, obviously, right?

513 "" (0) 03:45:06.279 --> 03:45:26.279 Until you hear what the number is for small publishers, which is 60 % 60 %, and I can attest to this both personally but also from knowing a lot of folks, this is the case, like the majority of trap service referral traffic have been reduced since 2023 for most small and independent travel publishers in my space. And.

514 "" (0) 03:45:26.279 --> 03:45:32.889 But I, you know, there's we can debate correlation correlation and causation and all of that, but I I think it's hard to argue that.

515 "" (0) 03:45:32.889 --> 03:45:49.959 AI isn't playing a role in that. And what's happening is, is that small publishers and small creators are not able to publish on the Open Web anymore. They're being pushed to publish on platforms, and, and the people who are publishing on the Open Web are largely very large.

516 "" (0) 03:45:49.959 --> 03:46:09.959 Corporations that have a lot of media properties and many of those corporations have publications that bring a lot of value to the web and I and I I think it's really important to talk a lot about news and journalistic organizations. I think they bring a lot of value to the web, but we need new publications. We need new websites, we need new creators to be able to publish the risk of the web is going to consolidate if we don't.

517 "" (0) 03:46:09.959 --> 03:46:15.099 If if we allow this situation to continue. So it's not and last thing I'll say is it's not just about.

518 "" (0) 03:46:15.099 --> 03:46:35.099 Sort of this, I think there's been some remarks that suggest that people contemplate this solely as sort of like a SEO for clicks like game. It's not just about that. 1st off, for many of small publishers, our search for for all traffic is so close to zero that like I haven't thought about SEO really in three years at this point because there's NO point we can.

519 "" (0) 03:46:35.099 --> 03:46:37.179 Such the minimust traffic from it.

520 "" (0) 03:46:37.179 --> 03:46:57.179 But I do care a lot about how, when my readers go to find my brand on whatever their preferred search mechanism is, I appear and i'll tell you a story that, and I hope I can do this without my my sort of the site that I work for, I'm being fully dragged into this, but this is just my personal experience. I spent December on.

521 "" (0) 03:46:57.179 --> 03:47:04.119 Going back through a content trip, through all of Peru I was working through the entire month, to create a series of.

522 "" (0) 03:47:04.119 --> 03:47:24.119 YouTube videos and written guides and short form videos, like there's like 50 pieces of content around Peru, one of which is an itinerary for Peru. And a lot of my YouTube videos have done pretty well. Lot of really good engagement and then people go in these we tell them in the videos like, hey, go search for search for my written guide, that's a companion.

523 "" (0) 03:47:24.119 --> 03:47:27.999 To this and when you go and search for it, you get an AI summary.

524 "" (0) 03:47:27.999 --> 03:47:47.999 That is, you know, if you search like the name of my site and prove itinerary, you get an AI summary, that's not my itinerary. And it's not, it won't even link to my itinerary, but it will show a photo of me that is extracted from my site, which gives it the veneerable look like it's my itinerary and the readers get confused. And sometimes the advice is actually.

525 "" (0) 03:47:47.999 --> 03:47:51.159 Bit of dangerous like I've seen where it will say.

526 "" (0) 03:47:51.159 --> 03:48:11.159 You know, you should go from Lima to Rainbow Mountain, which is super dangerous because Lima is at sea level and Rainbow Mountain is at 16000 ft. And that harms my brand. Like I don't want to be associated with a recommendation like that. And so even apart from like whether or not we get clicks, which we can't get clicks because they don't even put links apart from, you know, that.

527 "" (0) 03:48:11.159 --> 03:48:12.099 That image.

528 "" (0) 03:48:12.099 --> 03:48:32.099 You, you just wanna be able to control that. Recipe bloggers that I know have issues of what's called Frankenstein recipes where they mix together different, you can't mix recipes together, but these AI things and also by the way, it's more than just publishers like the and I I acknowledge that there are challenges with this, but like there was a very popular application last year.

529 "" (0) 03:48:32.099 --> 03:48:32.649 Sure.

530 "" (0) 03:48:32.649 --> 03:48:52.649 Where it became a trend to put her in the beginning where users would basically go out and find images of people they disagreed with online and because you have now like image to image models, they would, you know, they would upload the image and and and then that, that person would be notified basically and some governments ask laws about this and that platform.

531 "" (0) 03:48:52.649 --> 03:48:58.839 Has changed their policies, but there are reasons why people might want to express a preference with respect to.

532 "" (0) 03:48:58.839 --> 03:49:17.619 This usage that go beyond just any of this. And so I just think from my perspective, this is critical to the future of the web is also I think if we don't do this, like what are we going to come up with would be my question. Yeah. Thank you, Nick. So.

533 "" (0) 03:49:17.619 --> 03:49:36.639 I know before I said that we we don't want to try and collect use cases for all of AI, and so I'm reluctant to bring it up, but I I wonder if maybe a way to make a little progress here would be to try and differentiate and and and you know for for Brad, for Nate, for anybody else who, who, who wants this capability.

534 "" (0) 03:49:36.639 --> 03:49:56.639 You know, there, there's the case that, you know, a large search engine is using or or or a large chatbot is, is using the data as input into the context window, and that's one use case. Another is, is that I'm on my own machine, you know, summarizing or or using AI in some fashion to.

535 "" (0) 03:49:56.639 --> 03:50:16.599 Help me use data that I've acquired. And, and, and, you know, and then there are other use cases where maybe there's an open source project or an open source online service or there are a lot of different models for how AI uses and ingests data into its context window. And so, is, is your answer.

536 "" (0) 03:50:16.599 --> 03:50:36.599 The same for all of those that you want to express preferences for all of those or or is it targeted at, at, at a subset? So me personally, my preference would be NO AI almost always, just in life, but I re I think that other people might want to express more granular preferences with the.

537 "" (0) 03:50:36.599 --> 03:50:53.559 Respect to some of those things. I would say if the user question, like there, you have to then deal with like what do you do with open clock? Does that count as a user and like that, you know, where they have these like AI systems that can go out and scale and do things like collect photos and turn every person that you disagree with online, you know, in the beginning or whatever.

538 "" (0) 03:50:53.559 --> 03:51:13.559 I, you know, I I wouldn't I think there's probably an assumption going baked into that, that that's somehow good and I could be. I think that if you have a top level category that was another way to go about this is say there's a top level category, just know to AI and then you have an exception for search like we have now and that would accomplish what.

539 "" (0) 03:51:13.559 --> 03:51:31.629 Or we could have something like an AI summary, a mid level sort of approach would be kind of what we've had before, which is like AI output, AI summary grounding context windows, something that captures that second half of that math equation of the, of the input to the math equation. Like that is the logical approach here, and I think it's what the charter requires.

540 "" (0) 03:51:31.629 --> 03:51:51.629 The most narrow approach and one that I would prefer not to do because I I worry that it wouldn't be expensive enough, but what we could do is just that we currently have a proposed definition that's called non generative search, be logical, have generative search and to just cabinet into search. But of course then it becomes a question of what is search? And we do care. I I do care.

541 "" (0) 03:51:51.629 --> 03:52:11.629 About applications beyond just what I would call search engines, like e.g. being one of the big ones, like is that search I shared some examples in that blog post I put out, like it's really hard just looking at it to tell the difference, you know, which line these all up and be like, which one of these is search versus not search? So I think it's hard when you cabinet within the category of search, so I don't know. I mean.

542 "" (0) 03:52:11.629 --> 03:52:31.629 I'd be interested to hear what other people feel. I feel like any three of those, either the the big top level specific usage category or this more narrow one, could accomplish what I'm most interested in, I would prefer broader, but if narrower is what gets us to consensus then so I think like one of the.

543 "" (0) 03:52:31.629 --> 03:52:48.459 I'll do a short summary of something that happened before I'll let you go after. So one of the things that kind of was difficult with the substitute reviews categories like it can only be established after the fact, right? Like when you're when some like you you wouldn't know if like a summary is actually going to be substituted. I think that's the common like.

544 "" (0) 03:52:48.459 --> 03:53:08.459 Brad got when we last had the discussions, right? So do you have any like thoughts on how you go about addressing that, right? Like that I just want you to think through that like you know when you kind of talk about this a little bit further, but we have a time box discussion. I just want to put that on your mind because that was like one of the things that we couldn't get past, quickly. So.

545 "" (0) 03:53:08.459 --> 03:53:28.389 Echo, down to you. So I think there's a pretty like bright line here between a lot of the use cases we've been talking about where, you know, there's some agent that goes out and does a lot of things without really tight user supervision either you know either unilaterally when scraping or in rag use cases, and.

546 "" (0) 03:53:28.389 --> 03:53:44.319 Then the case where the user instructs the, the agent to take a specific action on a specific piece of content on a site. And, you know, I think I said earlier, you know, our job isn't to say legitimate and illegitimate, but like.

547 "" (0) 03:53:44.319 --> 03:54:04.319 I frankly find it pretty hard to believe that any agent is gonna care about any preference offered on the site that says do not take this content and just something with it unless it has some so degree of legal force. So, you know, if the agent, if I if I take my browser and I point it at, you know, at some site and I say summarize it or I say put this person in the beginning, like I expect my browser to do it.

548 "" (0) 03:54:04.319 --> 03:54:16.479 And not to care what the course what's on the site. Again, unless some real legal forced. So, I guess I'm skeptical of the value of those preferences, even though I suppose people should be allowed to say things that they wanna say.

549 "" (0) 03:54:16.479 --> 03:54:31.539 Thank you. Well, yeah, I think it might be useful to sort of in addition to the historical thing, also look at what others are doing currently, and there's sort of like a.

550 "" (0) 03:54:31.539 --> 03:54:46.959 Published an overview about this back in January but like there's at least, there's a number of, there's Leonard's AI inference, which is slightly different. The two other sort of like.

551 "" (0) 03:54:46.959 --> 03:55:06.959 Vocabulary style developed vocabularies that are out there and sort of like an offered to to publishers or anybody who wants to use them, help desk content signals and this really simple licensing consertium thing both have a AI input category which does seem to be.

552 "" (0) 03:55:06.959 --> 03:55:23.019 The relatively narrowly scoped towards the rack as for the delivery of search results or AI generated responses like I think they could be sharper on that sense, but there is.

553 "" (0) 03:55:23.019 --> 03:55:39.579 So both of them like vary in details of the wording, but they seem to be at least sort of things that approaches that we could consider, and that we haven't considered in the past. Historically.

554 "" (0) 03:55:39.579 --> 03:55:59.579 We had a AI use which was incredibly broad, and then we tried an AI output category that was, let's characterize it as complicated attempt to do this. And there is a, this, this much more straightforward AI input approach that is out there.

555 "" (0) 03:55:59.579 --> 03:56:16.479 And I think we somehow need to relate to also because both of those initiatives RSL and content signals, actually refer back to the IETF and express that they would like to be compatible with what comes out of here.

556 "" (0) 03:56:16.479 --> 03:56:36.479 Chris? Yeah, thank you. So, so so Suresh I just wanted to kind of talk about the point that you made earlier about the the substitution of use. Now, I think if I, when I read the text here, it, it speaks.

557 "" (0) 03:56:36.479 --> 03:56:38.319 Is that.

558 "" (0) 03:56:38.319 --> 03:56:57.039 You know, the th the application, it talks about summarization, it talks about aggregation, it talks about stylistic elements and so on. Like I I think there's perhaps a, a distinction here between the label that we've attached to it. Like we called it substitive use, which.

559 "" (0) 03:56:57.039 --> 03:57:14.079 You know, which is actually I think different to the, the way that it's described in the text, which is more about the, you know, the, the, the, the end result of of the processing, you know, I I agree with you like the whether whether it ends up being substanceive or not.

560 "" (0) 03:57:14.079 --> 03:57:29.169 Is kind of very context dependent and application dependent. But I think the language that we've got here in, in this proposal, you know, is, is actually really more talking about the, the, you know, the, the.

561 "" (0) 03:57:29.169 --> 03:57:45.189 Sort of the application of the processing. So, I would say if we need to change the title to make it more palatable to people, then I think we should do that. And I and I think this is a problem in in general, right? That we have.

562 "" (0) 03:57:45.189 --> 03:58:05.189 You know, we have the category labels and def, you know, and definitions that are, you know, that are, that express in more detail what the label kind of identifies. And so so those two things sort of need to be, you know, sort of.

563 "" (0) 03:58:07.929 --> 03:58:27.929 Coherent with each other and sort of consistent with each other, and I'm suggesting per you know perhaps, you know, that's not quite what we've, what we've got in this case. I I think, yeah, I agree with what people have said before like that there is absolutely the, the, the need for this category, but I'm, I'm wondering how we.

564 "" (0) 03:58:27.929 --> 03:58:48.419 Best go about defining this, whether it's in terms of, you know, act like, you know, textual summarization or image, video, audio style transfer, that kind of thing or whether we try to focus it on like the, the technical mechanisms that that happen in machine learning algorithms, you know, the processes that we.

565 "" (0) 03:58:48.419 --> 03:58:50.979 Of understand us.

566 "" (0) 03:58:50.979 --> 03:59:10.979 You know, retrieval augmentation or grounding and so on. Like, are these because those those sort of technical mechanisms end up being more like more generically applicable. And so if we're talking about, you know, whether it's it's text summarization or it's, you know.

567 "" (0) 03:59:10.979 --> 03:59:15.609 File transfer and things like that, then, you know, that.

568 "" (0) 03:59:15.609 --> 03:59:32.049 That that means that we, you know, we need to attempt to sort of enumerate more and more of those, those kind of application, sort of examples. So my my feeling is that we, you know, we should have something that, you know.

569 "" (0) 03:59:32.049 --> 03:59:52.049 People sort of tend to conceptualize this as, ok, well, we've got, you know, and I think the AI, the AI input category that we had before perhaps spoke to this was, you know, the use of content with an existing trained model that, that, you know, that, that produces outputs based on the content, you know, the, based on that content.

570 "" (0) 03:59:52.049 --> 04:00:11.049 So, so I'm I, I don't know, in my mind it feels fairly straightforward like as it's, you know, something that encompasses rag and, and grounding usage is really sort of, you know, what we're what we're after I think.

571 "" (0) 04:00:11.049 --> 04:00:31.049 Thanks Chris. I'm Brad. Yeah, I just Chris already made the point about the heading not, aligning necessarily with the language, the language being more specific and not requiring after the facts assessments. But I think maybe an approach.

572 "" (0) 04:00:31.049 --> 04:00:42.459 I'm just looking at what's been written on the list so far and the chat is that there's, I haven't heard anyone say we shouldn't have it because we don't need it. I think I've I've read everyone thinks.

573 "" (0) 04:00:42.459 --> 04:00:57.939 They have reservations because it's difficult to define and it's it's the concern is that it maybe too broad and it may encapsulate things that are activities that folks are not comfortable in in having a preference to restrict. So.

574 "" (0) 04:00:57.939 --> 04:01:17.469 I, I I I'm still hearing that like there's a, there's a there's there's some degree, a material that we have support for a concept of a preference, but, some folks are daunted by the challenge of of how to draft it and how to express it.

575 "" (0) 04:01:17.469 --> 04:01:33.849 I don't know if that's if that's that's an overly optimistic statement. I don't see how we, how we can walk away from this process and not, and not and not attempt to have.

576 "" (0) 04:01:33.849 --> 04:01:53.849 This as a, as a piece of a puzzle, like my my proposal was just a, was just a starting point. So, ok, so, sounds good right like so I think the, if you want to really go back to like where we had something like this, it's like version two of the draft, right? Like, you know, where we had like a aius category.

577 "" (0) 04:01:53.849 --> 04:01:57.579 Like yeah in there, right? So that was like a fairly.

578 "" (0) 04:01:57.579 --> 04:02:17.579 Simple definition. It didn't like torture, but we couldn't even get consensus on that, right? Like so I think the idea would be to come up with some text that actually has a chance of getting consensus on on the list, right? So, that I can, that was actually the act of using one of my assets as input to the trained AI ML model as part of the.

579 "" (0) 04:02:17.579 --> 04:02:21.969 Operation of that model. So that's the broadest, most sweeping.

580 "" (0) 04:02:21.969 --> 04:02:41.169 Type of language I've seen anywhere like basically use with a computer. Correct. Yeah. Correct. So that's, I wouldn't go back there, that's that. No, NO. I'm just saying like if you couldn't even like agree on like decide on yeah that we couldn't agree on that because that was extreme.

581 "" (0) 04:02:41.169 --> 04:03:01.169 No, but like the other direction is this, right, where we are trying to be pretty specific. So like we need to see like where we land on. Although I I do note that that kind of leans into I think what Kevin suggested yesterday, which is, you know, create preferences which are, you know, have to be violated. It's just, you know, but yeah, I I agree with you discovery.

582 "" (0) 04:03:01.169 --> 04:03:19.509 It's like saying don't use this data in any database. Don't use this data, yeah. That's it. Don't use this data with class of technology, yeah. Which raises the question why publish it in the 1st day. Nick.

583 "" (0) 04:03:19.509 --> 04:03:39.509 I'm here. Oh, you moved. Yeah. So I I don't think we should use a name technologies. I think this is what we've been moving away from. Sorry, i'll speak up. E.g., rag is one of those things that's not really well known. It could mean different things. I posted the.

584 "" (0) 04:03:39.509 --> 04:03:56.290 This in the chat, but like it used to mean having a vector database that would store additional data or a memory structure for an AI system, but now it sort of means any remote retrieval of information. So I I would.

585 "" (0) 04:03:56.290 --> 04:04:14.860 Caution against putting particular technical pieces in there and focus more on, I guess what users care about which is the use of the data and one line that you could draw would be whether or not data retrieved.

586 "" (0) 04:04:14.860 --> 04:04:34.860 Is used beyond the context in which it was retrieved. So if you, if you think of a user chat session that has a search application, it might fetch data from a website and that website might want that data to be incorporated into that chat session but not want it incorporated in.

587 "" (0) 04:04:34.860 --> 04:04:42.610 Into anything long term, whether that's training, whether that's put into a cash or whether that's put into a system that impacts.

588 "" (0) 04:04:42.610 --> 04:05:00.430 The output of that system for other users for for users outside of the context in which it was originally fetched. So I I I just I just caution that the use of of specifying specific technologies here rather than what we care about, which is the use of data.

589 "" (0) 04:05:03.520 --> 04:05:20.980 Okay, Kevin? I think I raised my hand a little while ago and the conversation feels like I just moved on. I was just sitting back and looking at the guy on the screen and I remember when Brad posted this to the list, I think it was after one of our previous interims.

590 "" (0) 04:05:20.980 --> 04:05:36.460 Where we'd sort of come around to realizing that this was the thing we were trying to describe, I think the problem here is that, that the definitions are too subjective.

591 "" (0) 04:05:36.460 --> 04:05:51.760 Right, like reduce the utility of, is a really subjective thing. So I, I don't, I see where this is trying to go and I appreciate that, but I think we we need more.

592 "" (0) 04:05:51.760 --> 04:06:08.560 Need less ambiguous ways to draw simpler lines. The whole point here, right, is that we want, we want somebody expressing a preference and consuming a preference to have a shared understanding of what that preference means.

593 "" (0) 04:06:08.560 --> 04:06:27.970 Yeah Nick? I think Max had his name. I think you were 1st. You chose, so I guess the way I contemplate it is is.

594 "" (0) 04:06:27.970 --> 04:06:46.690 You, if we say there's whatever we say with respect to like foundational base models, you then take that math equation and you take the assets and you append them to that math, you tokenize them and you append them to that math equation. Whatever that is, I call it context with nobody call it whatever you want.

595 "" (0) 04:06:46.690 --> 04:07:03.070 Like that's the thing that we should try to address. I definitely really like the idea of looking at what RSL and CloudFlare are doing, I think that they have really considered approaches. I'd be interested to hear if folks have concerns about those approaches so that we could try to design something that kind of makes sense there.

596 "" (0) 04:07:03.070 --> 04:07:23.070 I also, you know, I I I think, yeah, I just think that if there are other issues with an input based approach, like it would be good to hear those and and right now I'm just trying to come up with a list of concerns to try to address and like I get.

597 "" (0) 04:07:23.070 --> 04:07:43.070 The one about like specific user applications local, like, you know, what happens when they use it locally. To some extent I I feel like this is the flip of the discussion we had with the other categories, which is like we're starting to get into like normative uses of what we think the preferences should and should like how is it any different than all the discussions we just had about exceptions? I don't really understand why for this specific.

598 "" (0) 04:07:43.070 --> 04:07:47.260 The category would get so hung up on. Oh, it might sweep in all this.

599 "" (0) 04:07:47.260 --> 04:08:07.260 You know, socially good use, but then for the other category we're comfortable kind of not that we've resolved it, but anyway, I just think it's the same thing, so I don't think it, we should, we should be any further along on this. I think that this would be a lot easier to do as a whole, like to do search and train and this together because I think that they're all part.

600 "" (0) 04:08:07.260 --> 04:08:08.680 Of the same.

601 "" (0) 04:08:08.680 --> 04:08:28.680 So if there are other concerns beyond the this idea of like just being kind of truly broad and sweeping in other behavior with and specifically concerns with an input approach, we'd love to hear those. Thanks. I think at least like Kevin had a comment like that the this.

602 "" (0) 04:08:28.680 --> 04:08:30.280 Option of this is not.

603 "" (0) 04:08:30.280 --> 04:08:50.280 Concrete enough, right? So that's an Just take it into mind like time boxing to be clear like it's not like the this specific proposal I thought we started off with the basis that were like trying to come up with something new, so it's not necessarily about the substitute. I do think it's more, I said at the beginning, I think it's more than just it being substitutive I think it's also about control. Like I.

604 "" (0) 04:08:50.280 --> 04:08:55.450 I care primarily about controlling how my brand appears, which is different than substitutive.

605 "" (0) 04:08:55.450 --> 04:09:15.450 Right, like I I think the goal of this hour time box discussion is like for people to come up with like pixels, so if you have like some way of writing this and put it up for like people to look at it, I think that would be the best way to go forward. I mean the problem is is I'm not a machine learning engineer and what I'm saying if it's input based, it needs to be whatever that math equation is and so I don't even when I use terms like context.

606 "" (0) 04:09:15.450 --> 04:09:32.350 Do I get questions in the chat if I maybe wrong, so I I would love some, you know. I think maybe the best thing would be to probably pair you up with somebody who knows this thing and do try to express this together, right? Because I think that's something we've seen a lot of collaboration but kind of like, you know, in like groups of people who kind of have similar talks, maybe it's.

607 "" (0) 04:09:32.350 --> 04:09:47.620 Like interesting to kind of like cross pollinate a little bit like maybe like use it with Kevin and try to explain to him and like try to write something together like I'd be happy to be with you at the next break and see if I can help. Yeah, that'd be great. Yeah. It's pretty clear that we have a substantial number of folks who.

608 "" (0) 04:09:47.620 --> 04:10:04.540 Want something like this. And so the working group is going to be spending time on on evaluating proposals that they bring. If those are informed proposals that are that are not just, you know, one sided proposals I think it's a much better use of the working group's time if we can get that that input earlier rather than later in the process.

609 "" (0) 04:10:04.540 --> 04:10:19.900 Max? Yeah, I had a question for Paul, sort of at the end of the discussion was happening like 4 min ago ish, there was this comment on like.

610 "" (0) 04:10:19.900 --> 04:10:35.590 Like saying NO to, to all uses and data and I heard you say, then why publish? And I wanted to know whether that was an earnest question or just a quip. No, a quip. Just a quip. Okay. Thank you. Leonard.

611 "" (0) 04:10:35.590 --> 04:10:55.590 Yeah, I just wanted to comment that Nick's point about things like RSL or URL et cetera. You know, they absolutely serve a purpose. I believe though that they're orthogonal or complimentary to the work we're doing here. I think of them as extensions, and you can see that in.

612 "" (0) 04:10:55.590 --> 04:11:01.180 In fact, if you look at TDM rep, you look at both of them have mechanisms to say.

613 "" (0) 04:11:01.180 --> 04:11:17.080 And then solve, you know, so that where you can point out to the ODRL or RSL explanation, so I don't think that those are things that are gonna solve our problems for us, but there are definitely things we want to continue to think about. Thank you. Got it.

614 "" (0) 04:11:17.080 --> 04:11:34.060 I don't and not to sort of make Paul's arcastic comment the core of this, but I do think I have a little bit of that then why published sentiment and that sort of the core bargain of publishing is that you don't get to then later control the ideas and the information.

615 "" (0) 04:11:34.060 --> 04:11:49.090 In what you can published, and I think at least as they apply to individual users as opposed to technology companies, some of my concern about preferences about rag is to get to say I'm publishing this, but I want you to control the information in it.

616 "" (0) 04:11:49.090 --> 04:12:09.090 After publication. And that to me is problematic. Thank you. This is ideas versus expression kate. Can I address that? Just after Brad, like just hold on to your thoughts. Yeah, I mean I, I also think that one of the good pro quos of running a journalist and business or even a, you know, any publication.

617 "" (0) 04:12:09.090 --> 04:12:21.280 Is that in order to sustain that, you might need to sell advertising and actually earn money, which at this point in time comes from users visiting your sites and if that can't happen, then you can't do that anymore.

618 "" (0) 04:12:21.280 --> 04:12:37.600 So it's not just a question of making something public, which means that therefore everything you put out there is, is fair game. Even, even if it means extracting some of the more factual and less copyright protected elements of your site. Ultimately, if you're a.

619 "" (0) 04:12:37.600 --> 04:12:57.600 If you're a news organization and you're sending people out to war zones to get information to come back, you know, come to to to publish, some of that maybe factual, and yet you're hoping that by breaking the story people will come to your sites and that that will help sustain your business. So I don't think it's the.

620 "" (0) 04:12:57.600 --> 04:13:01.000 Because it's out there that there is NO cutoff.

621 "" (0) 04:13:01.000 --> 04:13:19.870 You know, financial like that not being able to control rag would materially disrupt and is materially disrupting as is evidenced by the number of small and local publications that are closing on a weekly basis.

622 "" (0) 04:13:19.870 --> 04:13:39.870 Thank you. Nick, did you wanna make up? Yeah, one thing to remember is it's not just about written texts, it's also about images. And I have a lot of images of myself all around the world that I published and I have other creators who work for me and we published their images and we care very much about how, how we're presented and we now have image to image models where you can change.

623 "" (0) 04:13:39.870 --> 04:13:57.220 Our appearance, and there's gonna be video to video models too, where when I upload a video guide, somebody could potentially alter that in all kinds of potentially problematic ways. So it's not just about the sort of text publishing model that was being addressed, so I think that's important to remember. Thank you.

624 "" (0) 04:13:57.220 --> 04:14:15.880 Kevin? Yeah, I have a question. I mean 1st off to just back up, like I, I feel like I agree we are not doing our job, there's not a preference that doesn't cover this. At the same time I'm wondering if when we go down this road.

625 "" (0) 04:14:15.880 --> 04:14:33.010 Do we need to be thinking about an exception depending on how it's displayed both for for both sides of the equation? I mean it seems like many publishers would actually and I think this will be increasingly true, want to appear.

626 "" (0) 04:14:33.010 --> 04:14:50.890 In generative search or in rag based results, if consumer behavior goes to always going to the chat window 1st rather than the internet if it is properly attributed. But right now if we're talking about just a broad saying NO to rag.

627 "" (0) 04:14:50.890 --> 04:15:10.420 This we wouldn't have that flexibility, which would seem useful to both sides of the table, but I I wanted to test that assumption. And I think like one more point I wanted to make to you Meredith right like from past discussions is that.

628 "" (0) 04:15:10.420 --> 04:15:30.420 Some of the publishers I think it was Chris like who brought it up in our discussion was that, the information you published might be free. In this case probably it is, and forgetting about like press things for a second, even though let's not ignore that, but the point was like some of the information was getting misrepresented, so imagine there's a summary that hasn't accurately.

629 "" (0) 04:15:30.420 --> 04:15:43.030 Summarize what you said, but it's still attributed to you. I think like Nate also brought up something like that, right? Like for the itinerary, so I think that is the other angle into this like that has been brought up before. So I just want to bring it up so that like.

630 "" (0) 04:15:43.030 --> 04:15:58.270 It's in your context for lack of a better term at this point. Thanks. To the, to that question so I, I just wanna I I don't have a direct answer, but I do wanna point out the context. There are already ways that you can, as a publisher.

631 "" (0) 04:15:58.270 --> 04:16:16.000 Through just robots standard block access to certain many of the chatbot applications where you could just block them entirely. Part of the problem is, is that unless you want to completely disappear from search, for some of the larger search engines you don't have a way.

632 "" (0) 04:16:16.000 --> 04:16:36.000 Too because they don't offer a granular control where you could opt out of the AI summary features but but remain in search. And so that's the fundamental problem here. And so regardless of what we do here, there are controls that you could use to say, I don't wanna appear in chat GPT, and in fact I use those controls. Those can be.

633 "" (0) 04:16:36.000 --> 04:16:53.920 The concern is, is that when's when when there's a timing between appearance and search, traditional search, and appearance in the the AI summaries, and there isn't a granular control there. I would add, we don't need the I the best path forward here would be for the.

634 "" (0) 04:16:53.920 --> 04:17:10.420 Large companies that that produce these products to simply provide those granular controls. And that could happen. I mean I know we're all individuals in this room here, but that isn't, that is, that could happen and that would be a better and faster path towards this, but there still will be ways to like.

635 "" (0) 04:17:10.420 --> 04:17:30.190 Some publishers maybe will wanna appear in these sort of chatbots and maybe some don't, and we should allow for that. We should allow for the expression I think it's most important the restrictive preference because that's really what we don't have. And if you want a different or permissive preference, those applications can offer controls. Like I think that that that's that's also important to remember.

636 "" (0) 04:17:30.190 --> 04:17:50.190 So we're, we're coming to the end of the time we had reserved for this. I see Krishna on the queue and anyone else, I think we want to prioritize anything else who hasn't had a chance to to speak, if, if, if you have anything to say, please put your hand up. Did you want to have it? Yeah, so there's like one thing I want to push back on a little bit nate, right? Like so when we started.

637 "" (0) 04:17:50.190 --> 04:18:10.190 Off on this work, right? So we did know that like, you know, Google, e.g., has like a way of you opting out, so it's like a bit more granular than this like being also had like a way of opting out of some of these summaries link. So the idea was like a lot of these providers, so you said the platforms can provide it and and each of the platforms is gonna have a different way of doing this and and some of the times like the.

638 "" (0) 04:18:10.190 --> 04:18:30.190 Publishers also came up with their own mechanism of how they want to opt out of things, right? So the idea of like kind of doing this work in the ITF is really not to do that because there's a combinative explosion of things that like everybody has like a different way of opting out of their thing, and then the publishers trying to do the same thing didn't work. So I just wanted to push back on the idea that like I would rather do that. It's the.

639 "" (0) 04:18:30.190 --> 04:18:49.510 The reason of this work existing is like to make sure like we can agree on something but so if you kind of pond that, I think then it has like much less value. So I just want to like throw that on there like I agree. I would rather that we accomplish this because this is more durable and also more neutral and and broadly applicable. I'm just saying that, you know, it isn't.

640 "" (0) 04:18:49.510 --> 04:19:09.510 There are, you know, if there are concerns that that more granularity or other things, we we've said that agreements outside of this, you know, could apply so there are other ways of doing that. I will caveat that by by saying if I can use privilege here, you know, there's a big difference between controlling or or or hinting the behavior of.

641 "" (0) 04:19:09.510 --> 04:19:19.750 Of, you know, a handful of large search engines and every use of AI everywhere. I mean the the latter is is the task that is causing us so many.

642 "" (0) 04:19:19.750 --> 04:19:39.370 For us to wrap ourselves into the circles because it's so broad. And why is that? Is that because we think that AI ought to be a social good that should be used in certain situations and going back and I, you know, I know we talked about accessibility in some of these things. And again, I feel like it's the same conversation for all these things or is it because it's something else? What I've heard is, is that NO.

643 "" (0) 04:19:39.370 --> 04:19:59.370 We don't know how AI is going to be used in the future. It is it is so dynamic and again it's it's, you know, you one can view it as just a computer science technique and and and so it is it's difficult to to to encompass all the possible uses, which is why the purpose based discussion I think was getting a lot of traction.

644 "" (0) 04:19:59.370 --> 04:20:16.720 Yeah. Thanks yeah so I just want to make a very simple observation over here. I totally hear the needs that are coming, right? Like very valid questions. I also want somebody to go back and look at the transcript of this meeting.

645 "" (0) 04:20:16.720 --> 04:20:33.310 And see and the number of times the word appear and display show up in the transcript. Every single example that I've heard so far in this particular part of the conversation relates to how things show up, how things appear, how things are displayed. Am I combined with something else?

646 "" (0) 04:20:33.310 --> 04:20:49.840 This is exactly what the display based preferences options allow us to do. So there is a proposal on the table related to that. So happens that it comes from me, but that's fine. But I just want to point out that what people say they need and what they're actually expressing in the transcripts if you go look at it.

647 "" (0) 04:20:49.840 --> 04:21:05.320 Are very much aligned to proposals that are already on the table, just putting it out there and I'll stop. Justin? Yeah, just one last hearing what Nate was saying, I think that.

648 "" (0) 04:21:05.320 --> 04:21:21.940 The risk of it being overbroad is that we remember that the preferences are preference. We don't have to follow them. We make them too broad, then they become a hammer, and then everyone will ignore them because they're just too broad. So I think we, there has to be a balance in the middle in order for us to get the most.

649 "" (0) 04:21:21.940 --> 04:21:41.940 Compliance that there's probably a better word to do. That's, that's a good point and I I I appreciate that and certainly we want people to use these. Christian, I'd love to talk to you about the display based thing. Sure. To preview sort of my concern every time I've thought about the display based thing.

650 "" (0) 04:21:41.940 --> 04:21:49.060 And in the past I've been supportive of one, is the problem of as small sites very often, we are like.

651 "" (0) 04:21:49.060 --> 04:22:07.660 The way that these AI search things work is they'll scrape or they'll I'm sorry, they'll input a hundred different, you know, assets, and then they'll only output links to three or five or whatever. And as a smaller independent site, we're very often in the 95 that are in the input window, but we don't get the links. And so.

652 "" (0) 04:22:07.660 --> 04:22:27.660 I the display based thing tends to not enable us to to express the preference that we don't wanna be contributing at all to that thing. And so we'll have to work through sort of happy to have a conversation with you offline about this. I would encourage you to go back and read the whole proposal as is expressed. It's not perfect. There are lots of places where it can change.

653 "" (0) 04:22:27.660 --> 04:22:28.270 Good.

654 "" (0) 04:22:28.270 --> 04:22:48.070 Things can evolve for the better but it there are some pretty **** good options over there for exactly the kinds of things that you're worried about. So it's worth considering if not right away at some point, this is kind of why Mark didn't kill it completely and kept it on the table and Mark and Suresh.

655 "" (0) 04:22:48.070 --> 04:23:08.070 They, when we had the conversation earlier, we didn't completely take it off the table exactly because of these kinds of scenarios that come up. So happy to discuss this. I hear the fashion, need to find a common growth. Okay, thank you. Do do you have any intention to update your, your proposal? Yeah, I'm this.

656 "" (0) 04:23:08.070 --> 04:23:21.700 This I'm using as a requirements gathering piece of exercise. This is why I actually look at the transcript quite carefully and I'm actually looking at what people are asking in terms of scenarios that they are describing as opposed to a very arbitrarily expressed want.

657 "" (0) 04:23:21.700 --> 04:23:39.760 Because ultimately I'm I'm a product designer and I want to make sure that we give a maximum options possible while keeping the product the floor because if you kill the products, then there is NO appearance anywhere, right? You don't have to worry about it. You solve the problem by killing the product. So it does, it needs to be a common ground somewhere, so I'm happy to go and look back at this.

658 "" (0) 04:23:39.760 --> 04:23:58.930 Okay, great. I think unless somebody else has like something to wrap up on this, like I think you're getting close to the end. Well we're not getting close to the end, but we're getting close to the the 6th end of this session this time box session. Anyone else?

659 "" (0) 04:23:58.930 --> 04:24:14.710 Anybody who hasn't spoken yet? Okay. So, the, the folks who are interested in this and the folks who are willing to help out, if, if you could collaborate at at this point.

660 "" (0) 04:24:14.710 --> 04:24:34.710 You know, we, we could have folks go off and and and do a separate thing and then bring something to the group that that's one model for this that works. I'm also, I think amenable to just doing the work, you know, if, if have a discussion on the mailing list and use it for if necessary, but you know the more that other folks are aware of things so they can point out concern.

661 "" (0) 04:24:34.710 --> 04:24:43.090 Learns the the more informed that work might be. But, but what we're really really looking for is again for shadowing the next time we meet.

662 "" (0) 04:24:43.090 --> 04:24:58.870 That that people have more developed proposals that we can use to, to, to seed the discussion and and make some progress much like we have with the, the search and the training terms. Anything else? Nope. Okay.

663 "" (0) 04:24:58.870 --> 04:25:18.670 Alright, let's go ahead and move back then. I think it was 1909. No, that was a full request. I'm looking for the discussion of training.

664 "" (0) 04:25:18.670 --> 04:25:35.170 Off there, what was that? Was it two oh one or two oh two or one is search.

665 "" (0) 04:25:35.170 --> 04:25:54.520 1909 is search as well. Those are the two ones. Oh, I think there was we're looking for 1908? Sure, 1908 at least. That's the one. Thank you very much.

666 "" (0) 04:25:54.520 --> 04:26:14.520 And so we had this proposal, and then we had this proposal. I think these are the two ones that were, there's nothing just a discussion. Okay. So have people had a chance to think about this more?

667 "" (0) 04:26:14.520 --> 04:26:34.900 The screen's having fun. Yeah, interesting mode. Right product. The internet? Yes yes it is. Yeah. Scroll up and down somewhat that's I'm not gonna move it anymore.

668 "" (0) 04:26:34.900 --> 04:26:52.510 I mean that's So this is what rough consents. It's rough. It looks really good on my screen. And eventually resolves. Thank you. It takes time. So I think the the core challenge here is.

669 "" (0) 04:26:52.510 --> 04:27:12.510 Whether the blanket definition, the one below applies, and that would be motivated by some of the things like we heard from, from Warren yesterday where attempting to create a carve out for simple applications, your ordinary leaf squares, your classifiers, the, the very simple things.

670 "" (0) 04:27:12.510 --> 04:27:29.050 Would be foolish because the classifier is a generator or it just you need to put a little bit of wrapping around it and suddenly you have general capabilities or whether we try to create that carve out by, by looking at.

671 "" (0) 04:27:29.050 --> 04:27:49.050 The generative capability that is inherent in foundation models and, more specialized models that are are specifically designed with the purpose of producing a particular type of content as opposed to doing some of those simpler tasks.

672 "" (0) 04:27:49.050 --> 04:28:04.900 Part of the discussion here was whether foundation model is still the term we want to be using. Let's leave aside the the name we choose for the moment because I think the definition is most important. I think to a large extent, we are really working at defining AI model.

673 "" (0) 04:28:04.900 --> 04:28:22.690 And that's probably the label that I would suggest we use in this context. It's just that AI is a term that has a meaning in people's minds that doesn't necessarily track with the definition that we're at. Or or or genders of the model or yeah.

674 "" (0) 04:28:22.690 --> 04:28:40.390 I, I think probably at this point the extra qualifier of generative doesn't really doesn't really carry a lot of value so I I think thanks Martin for for for this.

675 "" (0) 04:28:40.390 --> 04:29:00.390 Proposal, I think this text is pretty good. I agree with you there's not actually can find the foundation, however, so as you probably know, a number of a number of AI companies will let you fine tune their model, Your your their their models and your infrastructure and their infrastructure is like, you know, Google will let you fine tune Gemini even though you don't get access to like, you know.

676 "" (0) 04:29:00.390 --> 04:29:16.180 The Gemini waits and this wouldn't this would include that, right? Which I think is fine, but I just want to fly that that this is NO longer foundation model training, this is like any model training, which I think is fine, but I just want to have a flag that's. You're you're talking about the one at the bottom, correct?

677 "" (0) 04:29:16.180 --> 04:29:36.180 I think either one actually because what's on either the 1st two things on screen, I can't see if the same below it but if the refinement is the term that it, so I think you do need refinement because otherwise, you know, otherwise you couldn't cover fine tuning that happened inside.

678 "" (0) 04:29:36.180 --> 04:29:46.120 AI labs, but I'm just saying this also covers fine tuning that is done by users on on on AI lab on foundation models, which again I think is fine, but I just wanted to flag this with this says.

679 "" (0) 04:29:46.120 --> 04:30:05.470 Do you have any preference between the two, Anchor? I know I I think I think actually between these number one number two. Okay. Justin? Is the intent for this to only apply to models that generate content?

680 "" (0) 04:30:05.470 --> 04:30:20.590 As opposed to generate some form of output. So that, that is the difference that we're trying to to get between the two of them. The 1st one is an attempt to create a carve out such that models that.

681 "" (0) 04:30:20.590 --> 04:30:40.590 Don't generate content and not included. And this is where we're running into that challenge whereby those models with a little bit of wrapping, generate content NO matter. But I think that I think the argument would be that any model can generate content with wrapping.

682 "" (0) 04:30:40.590 --> 04:30:46.510 Because models in and of itself can't generate anything. They need a wrapper to do it.

683 "" (0) 04:30:46.510 --> 04:31:06.510 So I, I guess arguably either way, but I just I was trying to understand you're specifically trying to create this carver of models that generated. The 1st one is is yeah, it's the capability or the purpose or the intent.

684 "" (0) 04:31:06.510 --> 04:31:09.640 Yeah.

685 "" (0) 04:31:09.640 --> 04:31:29.290 Kevin? This sort of goes to what Mark was just saying. I think in my mind, the distinction between the capability of generating and the purpose or intent of generating is really important. When we say capability of generating, we're we're sectioning off a set of techniques.

686 "" (0) 04:31:29.290 --> 04:31:49.290 That produce models that could hypothetically be used to generate content, even if there is NO purpose or intent or actual use, of them for generating content. I'm just gonna float something out there. I I'm wondering whether it's it's intent that's important here or you.

687 "" (0) 04:31:49.290 --> 04:32:05.740 You know, would it make sense to say the specialized model that is restricted to, you know, only cases where it's not generating content? Just a thought. Is there an important distinction there in your mind?

688 "" (0) 04:32:05.740 --> 04:32:24.490 I think that that the the the purpose, the design purpose for a lot of people might cause questions, you know, about, you know, there there's, there are many things in the world that are designed for one purpose and used for other purposes, ok? And we've we've heard that concern from a couple of people yesterday and today.

689 "" (0) 04:32:24.490 --> 04:32:42.790 Think when I say that, I mean like the model would have associated restrictions that would be enforced through all of the normal compliance mechanisms that prevent us from Exactly. And and and so what I'm wondering if we, if we adopt this 1st one, there's probably gonna need to be some explanatory or or some examples.

690 "" (0) 04:32:42.790 --> 04:32:59.080 That that teased that aspect out Or we could just say used used gets there one?

691 "" (0) 04:32:59.080 --> 04:33:15.100 My name is Timod Robot. I like the 2nd one. I remain concerned about like a bait and switch intentional or unintentional, where.

692 "" (0) 04:33:15.100 --> 04:33:30.460 I think that the, there's too much distance between training, training a model and use of a model that especially especially with.

693 "" (0) 04:33:30.460 --> 04:33:47.290 With open source models and shared models, that there's NO NO effective way to know what a model will be used for, at, at the time of training.

694 "" (0) 04:33:47.290 --> 04:34:04.930 I agree with you to I think that's the concern, like I think we had like I don't know talked about like throwing up a open source model on HuggingFace and you don't know what people are gonna use it for after, right? Like, so I think that's like a valid concern. Like that somebody might.

695 "" (0) 04:34:04.930 --> 04:34:24.930 I'm not transfer this thing because in the model doesn't convey this perfect. If I were to argue the counter, I think you know we we have not been terribly concerned about enforcement till now. So, you know, the, the obvious outcome there would be if you're gonna throw a model up on a hung and face, you don't include anything that has this, you know, restrict.

696 "" (0) 04:34:25.450 --> 04:34:45.130 That's fair. Mike? I think there were other people before me, but my comment was not according to Webex, ok? My comment was along the similar lines to Mark so I think that that.

697 "" (0) 04:34:45.130 --> 04:35:02.140 Canada approaches it from the wrong direction, that we should not be thinking about this model can only be used for click and so, but that the model should carry the information that it was trained on content.

698 "" (0) 04:35:02.140 --> 04:35:22.140 That had consented to these uses and if you use it for something else, you are violating that consent just as if you were the one who trained the model for something else. So it flows downstream. I think maybe what you would draw from that is, is that in.

699 "" (0) 04:35:22.140 --> 04:35:35.350 In some cases models can, can have constraints placed upon them. If, if, you know, they're behind the firewall somewhere and the operator has exclusive control of it, then they can determine how it's used. Other models.

700 "" (0) 04:35:35.350 --> 04:35:51.550 If you throw them up on HuggingFace, you don't have those controls and so people need to behave accordingly. Or perhaps you've throw it up there but properly annotated. Yes, I have.

701 "" (0) 04:35:51.550 --> 04:36:07.000 Two concerns with the two definitions. The 1st one is the use of the term weight. It's very technique and and the definition of an AI model should be understood by content providers because they will express their.

702 "" (0) 04:36:07.000 --> 04:36:24.640 And and and this definition they won't they won't understand anything so the the world weight should be removed replaced by something a bit more or less technique. Can we use the word parameters?

703 "" (0) 04:36:24.640 --> 04:36:42.160 Would that work? Would that work better? Or if we can find a, a reference that we could use to explain this terminology perhaps. I'd be careful like that can be explained afterwards. There's like.

704 "" (0) 04:36:42.160 --> 04:37:02.140 The vocabulary and there can be an explanary here like let's not get bogged down like we have proposals here that we are turning around or since the beginning, like this language was on the table since the beginning of the working group. At some point we will need to say if we wanna go with A or B here.

705 "" (0) 04:37:02.140 --> 04:37:18.280 Otherwise we can come back in a couple of months and have this discussion again. And I I get your concern, but like that can be like there can be an explainer, there can be multiple explainers I think other people may at that point. Yep, thanks. Do we drop off?

706 "" (0) 04:37:18.280 --> 04:37:36.340 Hello, Tim robot. I think if we are going to go with the 1st one, and assume that like there's gonna be annotation following the model somehow that we've, we need to talk about what that.

707 "" (0) 04:37:36.340 --> 04:37:55.090 Annotation looks like. I think that's an, I would think it's in the attachment discussion, not in the vocabulary discussion, right? Like, I think at least like how Mike proposed it might be something that attached to the model afterwards, right? Like so I think it's not even relevant to this discussion now? I think so.

708 "" (0) 04:37:55.090 --> 04:38:11.980 Can we shelve that for a moment? Well I mean, but it's an argument that the concern will be addressed later, but if we're not kind of addressing it, you know, like it, it doesn't.

709 "" (0) 04:38:11.980 --> 04:38:31.980 Like have to be addressed later, right? Like, to say that like my concern that the use will change over time is irrelevant because like the model can be annotated, but if there's nothing about the vocabulary or the attachment that requires the model to be annotated.

710 "" (0) 04:38:31.980 --> 04:38:47.440 Right, but but you what you're doing is is I think this is getting awfully close to saying that we need to have a mechanism for enforceability of the preferences, which is out of scope for what we're doing. I mean, but it's just communicating the preference, right? Like the.

711 "" (0) 04:38:47.440 --> 04:39:07.440 If we're expecting the preference to be communicated across the lifespan of the model, so so I don't want to get into the weeds here, I don't agree with Mike. I I think that that the way that that would be effectively communicated is if you're gonna share something that could be potentially used for anything, you don't put anything in that that has training.

712 "" (0) 04:39:07.440 --> 04:39:10.900 It was now fine.

713 "" (0) 04:39:10.900 --> 04:39:30.900 Thank you. Yes. I think otherwise you're taking on the risk yourself, ok? So maybe I think let's call for preferences and then poll for objections. Does that make sense? Sure. Okay. So.

714 "" (0) 04:39:30.900 --> 04:39:44.560 So prefer, number one, prefer number two, NO preference to start with. Number one being the top one on the screen and number two being the 2nd one? It feels like all over again. You're in the ITPF.

715 "" (0) 04:39:44.560 --> 04:40:04.560 And Mark, just to be clear for the purposes of the poll, we're keeping the designs two and can. I think it's, it's again, we're we're looking for a starting point. The, the major difference here is do we have this split or not? And I think in both cases, we're gonna need to refine the language and you know you can treat those.

716 "" (0) 04:40:04.560 --> 04:40:06.700 Those friendly amendments I think at this point.

717 "" (0) 04:40:06.700 --> 04:40:25.240 I didn't hear any pushback on that kind of, at least considering that kind of adjustment. Cool, thank you. But yeah, this is not going to be final text in either case. Look good, by the way. Well I don't know how much this is Webex and how much this is the Cisco network, but it all goes down. Sorry?

718 "" (0) 04:40:25.240 --> 04:40:45.240 That's not sure to see a whole to get it. Oh. Yes. I think meet it goes made for right now. The poll echo, the poll hasn't started yet, are you yeah ok now but echo, I believe that what you do is there's a control somewhere.

719 "" (0) 04:40:45.240 --> 04:40:50.590 They're called apps or more panels, sorry, and then apps.

720 "" (0) 04:40:50.590 --> 04:41:09.550 And then provide the poll 1st and here it goes. Right. No preference is not an option, right? Yeah, NO preferences option. Sure. So I don't have that control. Yeah I don't either. It'll it'll appear in a moment with it.

721 "" (0) 04:41:09.550 --> 04:41:25.600 Here it is. Everyone should now have a slido. That's the easiest way. So Echo, it should be next to your participants icon.

722 "" (0) 04:41:25.600 --> 04:41:45.100 Off to the races. Did you fix something? This is behaving better for me today than it was yesterday. Yeah it was awful yesterday. I'm doing exactly the same thing I did yesterday. So I'm seeing the results now. Yeah. You enabled that then.

723 "" (0) 04:41:45.100 --> 04:42:00.100 Yesterday he didn't enable for us to see the results. 25 people. And what do we how many people do we have in the call? 41. Okay, so let's get the last few more people, oh 30 is good. This isn't worked yesterday.

724 "" (0) 04:42:00.100 --> 04:42:15.400 We have opinions. I like the fact that you like that. Last chance, if anybody else wants to put their preference in, it is just a preference that's not finding. It needs to be settling. If you're using it for transcellation.

725 "" (0) 04:42:15.400 --> 04:42:35.400 Okay, we seem to be settled there seems to be a fairly strong preference for number two, I think then the next thing that we would wanna ask and maybe not in a poll, but if anybody just wants to, you know, put their hand up, is there an objection to starting to number two? And if so.

726 "" (0) 04:42:35.400 --> 04:42:53.230 Characterize the concern. Can you put number two in the chat? Sure. The production put in the chat, that's ok. If we can find the chat, there it is.

727 "" (0) 04:42:53.230 --> 04:43:10.180 So, there's been a few, few friendly amendments that we we can consider. As I said, this is definitely a, you know, starting point. It is not the endpoint a model.

728 "" (0) 04:43:10.180 --> 04:43:28.000 So I'm not I'm not hearing anyone raising their hand or voice to to say NO that's not a good starting point. So let's let's use that one and we'll see how far we can get. So editors, why don't you start a pull request based upon that one if you haven't already?

729 "" (0) 04:43:28.000 --> 04:43:48.000 And start refining it and we'll take the PR to the working group and solicit feedback and eventually we'll try and do a consensus call on that. At the same time, we're gonna have to name this thing. So that's fine. I think it's AI. AI training? Yeah, alright, well we'll start with that and if people have concern.

730 "" (0) 04:43:48.000 --> 04:43:55.210 Concerns or or suggestions, we can talk through those as well.

731 "" (0) 04:43:55.210 --> 04:44:14.560 Anything else about this what the vocabulary term formerly known as foundation model training? Should we have a mozie? Okay. So it's 02:15.

732 "" (0) 04:44:14.560 --> 04:44:34.270 We haven't been away from lunch terribly long. I think we can just push through and take maybe an half an hour, we'll take another break. Yeah. So then that takes us to, the search term, which I think we have two PRs open for. One is 1909.

733 "" (0) 04:44:34.270 --> 04:44:53.380 And the other one is two oh one for memory? Yes. James, the text you put in, is this a proposal you're making or if you can come off mute and say something because it's kind of without context. I guys.

734 "" (0) 04:44:53.380 --> 04:45:10.390 Yeah, I just thought that that's just abbreviated, like multimodal is texting audio so why did you have to state it? I think that that's, we can get into fine edits and suggestions once we get a pull request under the list. I I'm sure that the language needs some tightening up.

735 "" (0) 04:45:10.390 --> 04:45:30.390 Sure ok thanks so let's see, I think the best way to look at these is using this view ish. What we had yesterday we put those two out and asked for comments like it was one substantial.

736 "" (0) 04:45:30.390 --> 04:45:37.690 Feedback on the two oh one, like I don't think we had subsequently any comments on the 199.

737 "" (0) 04:45:37.690 --> 04:45:57.690 So it might be more useful to to look at, there's a comment from Tyler from this morning which makes suggests a couple of amendments to like the two oh one language.

738 "" (0) 04:45:57.690 --> 04:46:13.060 Is that's on the PR? That's on the PR, I guess. Okay. I have to You know what, let's do this. I'm just gonna leave this here.

739 "" (0) 04:46:18.610 --> 04:46:36.460 I see. Okay. Oh, so the opening definition to direct users to the locations of those assets.

740 "" (0) 04:46:42.820 --> 04:46:58.540 And tighten the model exclusivity constraint issues. So these these all look like.

741 "" (0) 04:46:58.540 --> 04:47:15.430 Kind of tweaks to the language? That's super awkward to look at it in that form. Sure it is.

742 "" (0) 04:47:15.430 --> 04:47:35.430 Okay Leah, have you had a chance to look at these? No, not yet. I'm looking at it now.

743 "" (0) 04:47:53.710 --> 04:48:13.600 Okay, yeah, these all, all seem fairly.

744 "" (0) 04:48:13.600 --> 04:48:31.090 Are nearly editorial? I agree with Sorry, just to confirm you mean like Tyler and Martin's yes yeah I I agree with Martin's responses. Sorry, it's hard to hear you sorry I agree with Martin's responses verbatim was contested, so I've.

745 "" (0) 04:48:31.090 --> 04:48:48.850 By not having that, also same with the substitute of text it bears too close to legal assessments, and then on the exclusion piece.

746 "" (0) 04:48:48.850 --> 04:49:04.060 I'm kind of diagnostic to be honest. I'm sure there's ways to improve language. I'm also having a hard time reading like.

747 "" (0) 04:49:04.060 --> 04:49:24.060 As a version partly because I don't use GitHub much and partly because it's kind of like with national text now, so I'm not sure what the like current drop is actually. So if you if you go to the pull request to the files changed, I have done a pretty significant rewrite based on title's feedback, so I saw that which.

748 "" (0) 04:49:24.060 --> 04:49:42.400 That's good. And then you click this button right here to display the rich diff. I don't know if this is up to date now or not. Yeah, it is, ok. That gives you this view that I'm looking at here, which is for this particular change is probably the best way to look at it. Okay. Because then you just the tech, the new text is with the green bar on the left here. So it's essentially now like.

749 "" (0) 04:49:42.400 --> 04:50:01.270 A bunch of short paragraphs. So there there's the I think this is the current PR that was based on your original suggestion. It's not gonna scored a lot ultimately, but I think it's got it's got the same. Oh, it's got it, got it. Okay, sorry. I was reading it as all one.

750 "" (0) 04:50:01.270 --> 04:50:17.860 Yeah, the stuff with the red bar on the left is deleted So the 1st sentence doesn't change a lot further up and that's a bit. There we go.

751 "" (0) 04:50:17.860 --> 04:50:35.740 That that's largely just yeah. It might be worth, I think I saw someone make this comment somewhere, but we talk about asset in the singular, it might be worth seeing asset or assets. I think I fixed that, but.

752 "" (0) 04:50:35.740 --> 04:50:52.600 We can we can get there. It's not a I'm looking to make sure that we capture the substance of what people wanted from this one and something of the triangulation between.

753 "" (0) 04:50:52.600 --> 04:51:12.250 The different things, but I think this is the best that we've got so far. One, Yeah, except that I think it still mrs. some of the stuff with two oh two like today I was trying to book a hotel in Tokyo and I could not find anything useful because all of the results were showing up in Japanese.

754 "" (0) 04:51:12.250 --> 04:51:31.120 Having the translation for me was really useful, so you do say you know verbatim which is some square brackets or snippets can be displayed, but it's not, that's not actually useful to me, right? Like having snippets or experts in this is the contestant point I'm pointing to. Sure that's an open issue for that.

755 "" (0) 04:51:31.120 --> 04:51:50.560 Yep. Are you willing to, to leave this where it is? Understanding that that issue exists and we will continue to discuss that point, right? And again, at this point we're trying to choose a starting point for further work. We're not saying that this text is final in any ways. But I think that the contested bit is fairly fundamental to this whole second paragraph.

756 "" (0) 04:51:50.560 --> 04:52:09.610 Yeah, if we were to add that, I would remove the word verbatim. You'll notice it's in brackets and then and then add a sentence to that paragraph that says, doesn't include the use of assets to generate summaries and then, but it does include the.

757 "" (0) 04:52:09.610 --> 04:52:28.000 The ability to pro provide translations of short summaries of things. Interpretations of that. That's the tricky bit is that we have to work out what that takes. It would go right there. Yeah, we get into the slippery slope of how much can be like a conversion from.

758 "" (0) 04:52:28.000 --> 04:52:44.170 One language to another sure. But what if it is a more in depth sets of changes? That's exactly the challenge we need to. But the question is, are you ok with the understanding that if we come up with a solution to that problem?

759 "" (0) 04:52:44.170 --> 04:53:03.160 You would just go in there. I think so, but I think that we might be punting well but let's not face plan on this. The the the key decision now is which of the proposals do we want us to take forward you know let's look at the other one.

760 "" (0) 04:53:03.160 --> 04:53:19.840 Brad? Yeah, just that, that, that last sentence and sorry if I if this has there already been discussion about the last sentence that this admits training of AI models? We've discussed that and sorry I was.

761 "" (0) 04:53:19.840 --> 04:53:36.790 Meeting something. What was the sense of that? Cause that, that's troubling to me. It is it again this is this is one of those points that needs to be yeah confirmed. I think there are, there are, there's the point that Lauren raised.

762 "" (0) 04:53:36.790 --> 04:53:53.020 But I think the preponderance of the the discussion thus far has led to this particular sort of framing, which is as long as the the model is only used under these conditions.

763 "" (0) 04:53:53.020 --> 04:54:09.130 It's just the internal operations of the system and that's that's acceptable under this. But obviously some people are still uncomfortable with that. Yeah. Right. You know, I heard from Nate yesterday.

764 "" (0) 04:54:09.130 --> 04:54:29.130 If it were exclusively used for this application under the following conditions, that was what the sort of framing was. And this is where it's the wording is. And and and we don't want to trip up too much on exactly how it's expressed here and we can pick the language, you know, we might define special terms for search models or whatever. We'll get to that. It's more just is this the approach that we wanted to.

765 "" (0) 04:54:29.130 --> 04:54:48.640 I'm just not sure what the difference between a search model or any model that provides generative responses is anymore. So I think that where I understood there to be some con convergence was the understanding that there is going to be training in relation to the engine of this.

766 "" (0) 04:54:48.640 --> 04:55:07.870 The of the search part that is kind of backed into it. The way that it's described here kind of like envisages models sitting outside of that and now there is this kind of like tethering of that model to the use with this, you know, with with with the the search process. I think a.

767 "" (0) 04:55:07.870 --> 04:55:27.870 There's, there has to be a more elegant way to ascribe the activity that maybe training related or processing related, that if you want to be in search, then you've gotta be a part of that, but not necessarily to say, oh, by the way, train these models just as long as you only use them in connection with this product, which may evolve, by the way in the.

768 "" (0) 04:55:27.870 --> 04:55:40.780 Product is going to evolve in the ways that we will obviously not have control over so but but so it's in the list in this wording it's not tying it to the product so much it is, is specific uses.

769 "" (0) 04:55:40.780 --> 04:55:58.750 Only used as described, described above. Yeah. Yeah I in this section or what have what Editorial work TV editorial worked I I struggled with this one a little bit. I think that the key point is that.

770 "" (0) 04:55:58.750 --> 04:56:18.750 If someone in the woods, with NO observers trains and users of a model, that's none of our business at this point, because that model's not been taken somewhere else as an output. It's just an internal.

771 "" (0) 04:56:18.750 --> 04:56:35.620 Detail of how the system operates. And that's really the the the key sort of message there. Whether it's a model that's used or one of your I think search indexes on a database.

772 "" (0) 04:56:35.620 --> 04:56:53.410 It doesn't really matter. Whoever uses a database for web search, don't think that happens anymore if they did. We got a list.

773 "" (0) 04:56:53.410 --> 04:57:08.860 Do do we understand what this text is trying to do? You know, we've got a queue. Nate? So I appreciate the the effort here. I do think unfortunately it kind of brings a full circle back to the 1st concern I have specifically.

774 "" (0) 04:57:08.860 --> 04:57:25.690 The use of AI models as long as the outputs are only use to subscribed, the specific situation I'm worried about, like useful could include grounding and I'm concerned about that hundred where I'm in the 95, and so I could still be used because I'm not outputted at all.

775 "" (0) 04:57:25.690 --> 04:57:41.709 But I'm in the I'm the heart of the generation of that, of that answer. That's very often what happens to us is like very concrete and so it somehow needs to be, I think that's the 2nd paragraph here, which is.

776 "" (0) 04:57:41.709 --> 04:58:01.709 It can The model can only show 2nd sentence is good yeah and that's positional sentence of the model, yeah. Yeah. Yes. I think that the earlier approach, as I said yesterday of having like a core sentence and then inclusions and exclusions.

777 "" (0) 04:58:01.709 --> 04:58:09.669 Makes it a little easier because the hierarchy of, you're right, 2nd sentence of 2nd paragraph, like if I would, that would be the one I would claim, but I that.

778 "" (0) 04:58:09.669 --> 04:58:26.199 Final sentence could kind of imply and so I just maybe there's just a drafting thing. Yeah let's try to work on that. Let's have a look at the other one as well, but let's get through the queue and then we'll look at the other one. Chris? Yeah I.

779 "" (0) 04:58:26.199 --> 04:58:42.039 My initial thinking was about that final sentence was that, oh perhaps this is just an editorial thing, and then, you know, Brad's comment suggested to me that it's perhaps a bit more substantial that, you know, substantive than than purely editorial. So, so yeah, I I think.

780 "" (0) 04:58:42.039 --> 04:59:02.039 So my preference is that we still start with this as the basis, just to be clear, but I think I'd wanna see like a perhaps a, a tighter link between like the use and training of of AI models with the, like, with the internal process, you know, the internal processing of the, of the search application. So I think.

781 "" (0) 04:59:02.039 --> 04:59:09.789 As long as we can like tweak the wording to, to sort of bind those things tight, you know, more tightly together, I think this is.

782 "" (0) 04:59:09.789 --> 04:59:26.949 This is still preferable from my point of view. Thank you. Yeah, I, I think that this is very restrictive and would be difficult for a search provider to comply with.

783 "" (0) 04:59:26.949 --> 04:59:43.569 And and I think it's doing some things potentially unintentionally by trying to be so locked down. One example is the last sentence of the 1st sentence of the last paragraph, which says that internal processing.

784 "" (0) 04:59:43.569 --> 05:00:03.009 Performed on assets by the search application is out of scope of the preference, which means that it would be in the scope of a broader inference time category if there were one. And that could mean if the point of having a search category is to allow.

785 "" (0) 05:00:03.009 --> 05:00:23.009 You know, preference expressors or or content providers to opt out of a broad category and opt back into a search category. This would mean that all of the search engines internal processing is prohibited by the preferences, right? Please don't put things on the record that aren't adding evidence.

786 "" (0) 05:00:23.009 --> 05:00:28.929 This is we don't have an influence category. If we have an inference category, we'll have to fix this.

787 "" (0) 05:00:28.929 --> 05:00:44.169 But there's NO inference category at the moment and so within the architecture we have, that that's not a relevant concern. I think what Aaron's trying to say is in a world where this is this search category exists. Yes.

788 "" (0) 05:00:44.169 --> 05:01:04.169 And you say NO to everything, but yes to search, which is a use case that we've heard people say that they might be interested in. We don't have that capability though. How do you mean? We, you can say in our current capability, you can say NO to training and yes to search, but not NO to everything. We spent hours earlier today talking about.

789 "" (0) 05:01:04.169 --> 05:01:20.139 The potential of having that category. It was part of the discussion. Sure, it is not in the operative draft, but we're editing the operative draft. Do you, do you want to make this more complicated in anticipation of something that may may not exist? I want this category to work.

790 "" (0) 05:01:20.139 --> 05:01:36.009 And I don't think it does if you cannot perform the internal backend processing in order to do the display that it reports to. So I think some of this maybe.

791 "" (0) 05:01:36.009 --> 05:01:51.249 You can read does not apply in both directions. Right. Does this preference include doing the internal processing? Or does it say, it does not mean you are allowed to do internal processing?

792 "" (0) 05:01:51.249 --> 05:02:06.699 Which I think is I can fix that. No, I think it's just. Okay, I wasn't finished. Okay, I also think that.

793 "" (0) 05:02:06.699 --> 05:02:26.699 The continuation of that sentence conditions the potential internal processing on treating the asset according to the other restrictions which include a requirement to link to the asset. And so this could be an.

794 "" (0) 05:02:26.699 --> 05:02:46.509 Interpreted as prohibiting like ranking of the asset in a way that doesn't display it to the end user. I just to clarify it says like if you presented in the output, then it needs to get linked. I think that's what the restriction says. I don't think it's a blanket you should link to it. I I.

795 "" (0) 05:02:46.509 --> 05:03:04.179 I'm not sure that's what it says. It says this category applies where the presentation of an output includes a direct reference or linked. In search output. In search output. By the way, I read it the same way that Aaron does here for the simple reason because the word provided exists.

796 "" (0) 05:03:04.179 --> 05:03:21.069 And it is provided that the asset and the outputs of any processing is exclusively used by the according to the other restrictions in the section. So that and also this linking thing. So if you don't link, and if you don't adhere to all the other stuff.

797 "" (0) 05:03:21.069 --> 05:03:40.599 That could or could not exist, this condition would block the process, I think the words is and used could be interpreted as a positive requirement is kind of what you're saying. Yeah, I think is used according to the other restrictions in the section and the only restriction in the section is like.

798 "" (0) 05:03:40.599 --> 05:03:55.749 You must right right I see what you're saying, right? We can fix that. Yeah. Give me a second. I also have a concern with.

799 "" (0) 05:03:55.749 --> 05:04:12.069 The reference to summaries, I I would like a definition of that. Yep. If we are going to move forward with this definition. That's been pretty clear from, from previous discussion, yeah.

800 "" (0) 05:04:12.069 --> 05:04:28.449 Okay, let's look at the other one briefly, make sure we understand it, then we'll take a break and come back and try and figure out where to go from. So this is the other one, which almost fits in the screen. Let me one. Oh, please.

801 "" (0) 05:04:28.449 --> 05:04:43.899 I think this 1909, right? Yeah, this is 1909. So here we are.

802 "" (0) 05:04:50.379 --> 05:05:10.379 Two of those, not the summary thing. Okay.

803 "" (0) 05:05:10.379 --> 05:05:27.939 Okay, any concerns, questions? Aaron is your hand back up or NO, that was sorry. Chris?

804 "" (0) 05:05:27.939 --> 05:05:47.939 But yeah, my hand was actually on the previous point, on the, yeah, the previous proposal, which is, the word ranking was, was in Leah's previous text and hasn't been carried forward into, into that pull request. So that I I think ranking is something that you know that we could add in as part of the description of that, the internal.

805 "" (0) 05:05:47.939 --> 05:05:50.049 Processing. However.

806 "" (0) 05:05:50.049 --> 05:06:09.699 On on this particular right yeah so that was just the point on on the previous one. On this one, I think what we need to do is because we're, I'm focusing on the the verbatim part now.

807 "" (0) 05:06:09.699 --> 05:06:24.939 So this is one of the points of differences between the other proposal. My concern here is that when we is I think we need more clarity around sentence level text, if the intent here is, you know, limited.

808 "" (0) 05:06:24.939 --> 05:06:42.879 Generative output to the extent, you know, that's limited to like a single sentence, then that's fine, but you know paragraphs are made of sentences and, and so sentence level text.

809 "" (0) 05:06:42.879 --> 05:06:59.859 Doesn't I think constrain it enough because you could apply that to, you know, paragraphs and then all pages and so on. So I I think we'd want to see a bit more of a, a clearer limit set around what we mean by sentence level.

810 "" (0) 05:06:59.859 --> 05:07:19.299 Could I jump in really quick? Sure. I totally agree with that. I didn't want to just say a sentence because sentences can be very long and multi clausal and, so sentence level was my attempt to get at like things that are about the length of a normal sentence ish, but I.

811 "" (0) 05:07:19.299 --> 05:07:35.079 Like, very much welcome more concreteness there. I think it is needed. I, I don't think such thing exists, like I think yours is probably the best description of what you want. Yeah, which we could also just say something.

812 "" (0) 05:07:35.079 --> 05:07:55.079 Not perfect and then give a bunch of definition for it in the definition. Yeah, sentence one. So I started doing this in chat. One of the one of my concerns is the images of any sites. So I just posted a.

813 "" (0) 05:07:55.079 --> 05:08:13.149 The search that I happened to do on Google and I figured out which page I wanted to go to based on the fact that there was an image of it which was resized from an image on the page. I suspect that there was likely some AI involved in a fair bit of image resizing and.

814 "" (0) 05:08:13.149 --> 05:08:30.000 Regeneration. And this sounds like that rules that out of scope like resizing an image these days often involves some sort of AI stuff to decide which are useful pixels to actually be at which aren't. And.

815 "" (0) 05:08:31.410 --> 05:08:51.410 That could be viewed at the generative wording in this. It's already giving me a depth stare, so i'll stop working. So it says images of any size. Yep, so that basically rules out the the use of images period. Yep. That's, that's my concern. Yeah, I think that's probably But I don't know how we say like images that.

816 "" (0) 05:08:51.410 --> 05:09:02.370 Not obviously something you can use for other things like how do you work with this? Does resizing requirement? Generally, often. It doesn't require though, but.

817 "" (0) 05:09:02.370 --> 05:09:19.380 Do you get better, do you get better results if you know that you should ignore, you know, the leaves in the background and the trees? And, and to play devil's advocate, audio visual content could perhaps means an assist of technology. Yeah.

818 "" (0) 05:09:22.830 --> 05:09:42.830 However I pointed out an issue I don't have suggested text, so you're welcome Markin. Yeah, I know what to do with that. So, I'll raise this is again the same issue I I mentioned before, which is what happens when you're among the 95.

819 "" (0) 05:09:42.830 --> 05:09:55.560 Five that don't get a link, but you are part of the processing in the context window. The way this is framed, I think this one quite clearly has an issue there because it, it applies to use of an asset in search results.

820 "" (0) 05:09:55.560 --> 05:10:14.670 So that's, it only applies when use of asset and use search results, and then search result is defined as the links and associated content presented to users. So if I'm not given a link and presented to users, then this doesn't this doesn't say anything, and so you could then.

821 "" (0) 05:10:14.670 --> 05:10:32.850 You know, you could still do that and but then this you know or another another body might like consider this to sort of have been consent to that. And so it's we're back to my same concern, which is it needs to be explicit. You cannot, like, I I don't want to be in that context window if I say yes to this.

822 "" (0) 05:10:32.850 --> 05:10:52.850 On Absent of reference, you know, in the output, like what happens to you who are in who are in the input but not in the output? That's the question. I I I don't know if, if, if, I personally agree with you, but I think your comment does for me point out that that 1st sentence on the screen is.

823 "" (0) 05:10:52.850 --> 05:11:09.540 Very selective and ambiguous about what's in scope and out of scope here, and I think that leads to your problem and also other readings that could be really problematic change, I think. Well, that was to keep it as a search category, right? Not a generic category.

824 "" (0) 05:11:09.540 --> 05:11:29.540 Right, but, but it's about use of the added asset in search results, which is a very specific thing. It's not by use of the asset by a search application. Right. Yeah. Yeah. Yeah. But I think as Martin just pointed out.

825 "" (0) 05:11:29.540 --> 05:11:49.410 We're not talking about things other than like the fairly narrow traditional search result type category at this moment. Martin? Yeah, so, just on Warren's point about images, I I think this is.

826 "" (0) 05:11:49.410 --> 05:12:09.410 Because this is tied to generation, and a lower cases that we're talking about, there will be generative capabilities involved in selecting which image might be presented next to a particular search result, even selecting what part of an image to crop to so that it.

827 "" (0) 05:12:09.410 --> 05:12:28.050 Becomes, but that's not generation. That's back to excepts and snippets. But there's a generic model involved in the pro in the internal processing of these things, but we don't care how the decision is made to select a snippet but then it generates the image to resplit, not.

828 "" (0) 05:12:28.050 --> 05:12:44.880 Really? No. It's just taking the existing image that's there and choosing what part of it to present, and there's a process that happens, but that's not generation of anything substantially new.

829 "" (0) 05:12:44.880 --> 05:13:02.339 Do you, do you imagine that you would get hallucinated images next to these articles? Hopefully not, but I I don't know if it is just take image and display only prop of it.

830 "" (0) 05:13:02.339 --> 05:13:18.119 I don't know if there let me let me try and find an example. If you have examples of where, where there is concretely substantial generation involved in the images, I'd be somewhat surprised given the cost of creating one of those, but.

831 "" (0) 05:13:18.119 --> 05:13:34.049 Yeah, we also don't want to be locking this to only what exists now. I understand but we have to we have to base what we have to. Yeah, you know, you've been doing a huge chunk of the text generation and people are like.

832 "" (0) 05:13:34.049 --> 05:13:49.649 I don't like this to do it differently. Yeah. Bring me a rock. Yeah, the bring me rock thing is is a big problem, yeah. Thanks. Justin and robot, you do have the last words before going for a break. I understand.

833 "" (0) 05:13:49.649 --> 05:14:09.029 Where you're coming from with the images, I think it's it's a little bit vague, so I think some clarity would I think everyone has some consensus. On the sentence level text commented the the sentence above states not used in the generation of substantial quantities, substantial quantities being sentence level text.

834 "" (0) 05:14:09.029 --> 05:14:27.269 That speaks to generative summary, which I know was was a term we were going to talk to, so I'm just confused as to where generative summaries plays a role because I think there is, there is a wish that some wish to be in the search, but not have any generative summaries applied.

835 "" (0) 05:14:27.269 --> 05:14:47.269 And I'm just wondering how that how that. So Justin I think there was like a clear distinction between excepts and summaries. So I think that should address your concern. But it does. There's, I just want to make sure that it's clear inside in this text verbatim excepts means something that was taken off of the page.

836 "" (0) 05:14:47.269 --> 05:14:50.999 Okay, and as long as the sentence level text.

837 "" (0) 05:14:50.999 --> 05:15:08.279 Covers the gendered summary piece to say if I say I don't want gendered summaries, then I don't. As long as I'm understanding. Hello? 1st regarding images.

838 "" (0) 05:15:08.279 --> 05:15:25.739 I don't know why we would need to treat those differently in text as far as like about like excerpts etc. I would also be very surprised if there was a generation or if anyone wanted.

839 "" (0) 05:15:25.739 --> 05:15:43.049 Image generation and they're representing their site. And then, I just wanted to chime in that I personally find the two oh one proposal to be an easier starting point for me.

840 "" (0) 05:15:43.049 --> 05:16:00.209 Thank you. Alright, so let's take a break until three, and then we'll come back and we'll, we'll try and figure out which one of these we're gonna use as a starting point to go forward. And we'll have lots of caveats and context around that to make sure that we make an informed decision.

841 "" (0) 05:16:00.209 --> 05:16:17.189 So, let's be back at three. Three. All right. Give them permission. Sure. You can.

842 "" (0) 05:33:08.329 --> 05:33:28.329 Yeah, yeah.

843 "" (0) 05:33:28.329 --> 05:33:48.329 The expressive part of the reason for the people who are remote people are still trickling in. Now training model, but then he doesn't want to see his stuff.

844 "" (0) 05:33:48.329 --> 05:33:50.819 I know I'm late, but.

845 "" (0) 05:35:42.839 --> 05:35:57.929 Okay let's let John get started again. John. He's calling names. Daniel.

846 "" (0) 05:35:57.929 --> 05:36:17.929 I'm dirking coffee, but he's calling names. Sure. What are you doing? In which order? So we'd like to do.

847 "" (0) 05:36:17.929 --> 05:36:35.309 Folks, we'd like to do, I see a few empty chairs, but hopefully they'll gravitate back in. I'd like to do what we did before just to try and choose one of these as as the basis for further work. And again, it's choosing a starting point. It is not choosing an endpoint.

848 "" (0) 05:36:35.309 --> 05:36:55.309 There are many different things that will probably change about these techs. We'll raise issues against them. We may take input or or or snippets from the one, oh geez snippets from the one that we don't choose, and incorporate them. What we choose may end up indeed looking like the one that we didn't choose at the end after.

849 "" (0) 05:36:55.309 --> 05:37:08.879 Substantive discussion. So please don't over rotate on this. It is, it is not something that you need to win desperately. So what we'll do is we'll ask for a preference poll.

850 "" (0) 05:37:08.879 --> 05:37:28.879 Number 1909 is this one right here and I'm not gonna associate these with anyone's name because it's not about that. This is number 1909, and this is number two oh one and the 3rd option will be NO preference. And then we'll do what we did before, which would ask for objections if we have a clear preference and talk through.

851 "" (0) 05:37:28.879 --> 05:37:47.729 Do that if necessary. So Goalmaster, please create the poll. The one on the screen now without the hover, is two oh one. This is two oh one yes? And one with the long list of bullets is 1909.

852 "" (0) 05:37:47.729 --> 05:38:06.179 This one which will appear hopefully before the heat depth of the universe is 1909. You click refresh I will click refresh because like the oh great.

853 "" (0) 05:38:06.179 --> 05:38:23.849 Here we go. That is 1909. Oh, there it is. Yeah, so you're creating a poll? Yep. All right, the poll is launching.

854 "" (0) 05:38:23.849 --> 05:38:42.749 Go forth and poll, a coffee. I like what.

855 "" (0) 05:38:42.749 --> 05:38:59.159 I'm glad I can amuse you Martin. By whatever changed I did, I don't know what changed I did in slido. It's good. The fact that only you could see the answers before was good.

856 "" (0) 05:38:59.159 --> 05:39:14.489 Although it does influence how people it does. Oh yeah, that's fine. You're right. You're right. Unless you're looking. Yeah don't.

857 "" (0) 05:39:14.489 --> 05:39:33.779 Okay, 24 out of 40 more people if you want to express your preference for the preference? It's preferences all the way down. You got a question? Just a question in the direction of preferences.

858 "" (0) 05:39:33.779 --> 05:39:52.949 Now I know I need to watch you. You have your hand up? Give me to a buck? No, not again. Sorry. Okay, just a few more moments if you haven't.

859 "" (0) 05:39:52.949 --> 05:40:12.119 Expressed your preference, please do so. It's looking like well it does not. Okay, let's call it that's 34 oh 36. Come on people.

860 "" (0) 05:40:12.119 --> 05:40:28.529 It's not conclusive or checkbody's changed their preference. Not terribly conclusive, two oh one has a lead, but not a great one. It's closely followed by 1909 and the ever.

861 "" (0) 05:40:28.529 --> 05:40:45.629 Helpful, NO preference. Should I know preference, take it out. Well, let let's just ask the question. Two oh one has a slight lead. Does anyone have an objection? They think that that you know it's gonna substantially set us back if we start from number two oh one.

862 "" (0) 05:40:45.629 --> 05:41:02.699 Knowing that, as I said before, everything is still on the table, we're still gonna work through the issues and make sure that we refine the text to get it to the place we need it to be. Yes. I put this in the, in a comment one two oh one but I think it's a good starting place if we take in that sentence about.

863 "" (0) 05:41:02.699 --> 05:41:22.699 Translation text to speech just as a placeholder, like even if we bracket it, so that we're starting to at least represents the ideas from both. So just give it that kind of equity between the two. Put a textual placeholder in for that list. I I'll do that right now I've got it. Okay. Well, so assume that it has a placeholder for the.

864 "" (0) 05:41:22.699 --> 05:41:40.889 Accessibility stuff and knowing that of course that's still something we need to discuss, but just to give it that kind of a quality, if that changes your vote. Oh, darren. Yeah. Okay. So any, any objection to using two oh one as as the the basis of our continuing discussion?

865 "" (0) 05:41:44.849 --> 05:42:03.389 Okay, let's go ahead and do that. So, editors go ahead and and continue to work with that PR. Let's take the discussion to the list. Let's figure out what needs to still be discussed in that issue or or in that PR and and and move forward with that.

866 "" (0) 05:42:03.389 --> 05:42:23.389 But don't merge it yet, obviously. So, that leaves us in a place where we have the the basis for further discussion on the training term, the basis for further discussion on the search term.

867 "" (0) 05:42:23.389 --> 05:42:39.719 I think the, you know, we have about an hour left today and then we have tomorrow morning. The things that seem like they could benefit from more discussion are going back to the three.two discussion, which we talked about earlier.

868 "" (0) 05:42:39.719 --> 05:42:58.409 And, and, and continuing that and then continuing the refinement of these texts both for training and for search. Does that make sense? Is as is there anything else we need to have on the agenda for the rest of the meeting?

869 "" (0) 05:42:58.409 --> 05:43:15.719 One of the thing we will discuss, I think tomorrow and and If Does anyone need to leave substantially before, let's call it? 1230 or so tomorrow? I think we do have at least a few people who will not be here at all. We have people who aren't here tomorrow at all, one.

870 "" (0) 05:43:15.719 --> 05:43:30.809 Okay, well maybe we'll we'll talk about it today then. Suresh and I've been talking and we've been talking to Mike. We're we're generally pleased with the, the, the progress we've made in this meeting. We we had concerns coming to this meeting that if we didn't make progress, we would.

871 "" (0) 05:43:30.809 --> 05:43:46.949 Few questioning we're doing here, but but so far we're, we're, we're pretty happy with with people, you know, listening to other viewpoints and trying to incorporate them and and compromising, so that's great. And so it's pretty obvious we need to.

872 "" (0) 05:43:46.949 --> 05:44:03.359 Have more discussions to continue this. We would appreciate it if the work group can make progress asynchronously on the mailing list. However, past experience is showing that that's difficult for us for a variety of reasons. So what we're thinking about at this point.

873 "" (0) 05:44:03.359 --> 05:44:23.359 He is holding another on owned interim meeting roughly in June, so kind of in a month and a half ish. And then of course we have the Vienna ITF meeting. The most we can probably request in that meeting is two, 2 h slots, so that's 4 h.

874 "" (0) 05:44:23.359 --> 05:44:43.359 We'll have some discussions there, but like everything we won't be making decisions, and at least part of that time we'll be taking whatever we are able to come up with back to the greater ITF community and socializing that. And then it's it's likely we'll need another meeting like this after that to kind of close up.

875 "" (0) 05:44:43.359 --> 05:44:58.799 Both what we've talked about here and also if we have proposals for new additions to work through those. And so we're talking about another meeting like this, kind of in the late August, early September timeframe. And we're thinking the thinking is we'll go back to Europe.

876 "" (0) 05:44:58.799 --> 05:45:15.659 So that's none of that's official yet, we haven't scheduled it, so please don't book any travel obviously. But that's what we're thinking about if folks have dates that are better or worse than those time frames in late August and early September, please let Suresh and I know.

877 "" (0) 05:45:15.659 --> 05:45:31.859 If you have a facility that you can offer to host a meeting like this, keeping in mind the size of the group and the need for things like a network and the need to have, you know, NO NDA to get into a building, that sort of thing, please come talk to us. I think we, you know.

878 "" (0) 05:45:31.859 --> 05:45:48.329 From the natural thing to do is is to fall back to using CloudFlare and London because we know there's a space there that works. But if other folks have spaces that could be used, come and tell us. Yeah, so that's kind of the plan going forward.

879 "" (0) 05:45:48.329 --> 05:46:08.329 And, and, and hopefully we'll get to a place where at the next one of these meetings we can talk about you know what it's gonna take to actually ship the specifications. We haven't talked at all about vocabulary yet in this meeting. Sorry, not vocabulary, attachment. We we can, I think there's a couple of open issues on attachment, but I'm not sure.

880 "" (0) 05:46:08.329 --> 05:46:12.029 Or you know how productive and if if people are prepared for that.

881 "" (0) 05:46:12.029 --> 05:46:29.189 So let's continue the duplicate discussions for now and see where we get with it. Can I just do your, do your planning and to the lack of discussion on the mailing list? Like I think also there weren't very many consequential actions.

882 "" (0) 05:46:29.189 --> 05:46:44.250 On the maining list in the past couple of months I do have a feeling to get a sentiment in how how far the progress that I agree we've made here wholes, it would be important to put like.

883 "" (0) 05:46:44.250 --> 05:47:04.250 Whatever comes out of these two categories like our editing on these two categories to some form of consensus so very much so. I, you know, depending on how much we can discuss around refining these proposals in the next, you know, hour today and tomorrow morning, and then what we can get done on the mailing list, I I.

884 "" (0) 05:47:04.250 --> 05:47:19.380 I think we can see putting out consensus calls on the the training and search terms in the not too distant future. Now it maybe we need to have further discussions at that online only interim to get there if if substantial issues pop up, but.

885 "" (0) 05:47:19.380 --> 05:47:39.380 I'm I'm really hoping that people can engage, you know, now that we have concrete proposals on the mailing list and say, ok, well, I have this issue with the text, can we talk through this and do that on the mailing list rather than waiting till the next time we're all in the same room again? Because if we keep on only making progress in rooms like this, this is going to take.

886 "" (0) 05:47:39.380 --> 05:47:59.380 For a while. Yep. The other alternative is, is if we can't do that, we could move to a much more aggressive face to face meeting schedule, and, and instead of doing it online in June we do it face to face in June, but that creates issues around equity where people who can't come to these meetings get disadvantaged, it creates more burden for people who have other commitments and I especially.

887 "" (0) 05:47:59.380 --> 05:48:05.700 You see our editors in mind with that. So I would prefer that we make progress in the mail list where we're, we're able to.

888 "" (0) 05:48:05.700 --> 05:48:25.700 Yeah, and the other thing would be like, before the next interim to have some kind of description of the use stuff because like you know Brad, Nate, you've talked about stuff, I think it would be good to kind of have a proposal on the table before the next online interim, because like I think if you get too close to it before people have.

889 "" (0) 05:48:25.700 --> 05:48:31.620 We're not gonna make progress on that, so that will be good to do it that way. Thank you.

890 "" (0) 05:48:31.620 --> 05:48:51.620 Would you schedule the topic of the attachment mechanisms for that August September ish meeting? I'm thinking off the top of my head we're gonna start we wanna start talking to about attachments pretty soon again. So now that we, you know, we're.

891 "" (0) 05:48:51.620 --> 05:48:59.250 Hopefully get to the point where we can nail these terms down the next obvious thing and and the other parts of the vocabulary spec, I think we would devote.

892 "" (0) 05:48:59.250 --> 05:49:18.600 Perhaps part of that online interim in June to restarting the discussions around attachment and familiarizing people with them start thinking about the issues that are outstanding in that spec and then in in I think especially in in Vienna might be a good time to talk about attachment because.

893 "" (0) 05:49:18.600 --> 05:49:38.600 There are more people at the ITF meetings with perhaps relevant expertise in terms of how things are attached to protocols. So we'll have more discussion there and then of course in in the the next hybrid meeting as well.

894 "" (0) 05:49:40.080 --> 05:49:59.790 So, I think we're gonna go for two oh one there. We didn't hear any objections. Do we want to just since since we've got this paged in, do we wanna continue the discussion? Oops, can we, can we dig in on this for better question?

895 "" (0) 05:49:59.790 --> 05:50:19.790 Sure, we need an answer to it. I know it's not easy, but, I'm trying to take care of a suggestion and it's pretty hard to write. What I have right in front of me is non substantive changes to the.

896 "" (0) 05:50:19.790 --> 05:50:25.560 Presentation of titles, excepts or snippets from assets is included in the definition.

897 "" (0) 05:50:25.560 --> 05:50:41.100 That would, that would mean it's possible. That would admit the possibility of a translation of the text that's presented in the snippet, as opposed to just verbatim copies of the snippet and would allow for.

898 "" (0) 05:50:41.100 --> 05:50:58.110 Transcription of the text from a video or audio message or text to speech, text to speech for those people who are unable to read.

899 "" (0) 05:50:58.110 --> 05:51:16.680 So, these are basic accessibility affordances that are part of many applications already. The question is whether or not the risk of, malicious paraphrasing is, is unacceptable.

900 "" (0) 05:51:16.680 --> 05:51:32.580 Just wondering if this, does this belong, does this belong here or is this not part of the discussion that we were having earlier?

901 "" (0) 05:51:32.580 --> 05:51:49.650 About, you know, areas in which, you know, you may decide not to respect the preference. I mean, doesn't, doesn't, I mean it it I mean it it it clearly overlaps with some of what.

902 "" (0) 05:51:49.650 --> 05:52:07.290 Should suggested, but now we're we're just having that conversation in relation to a specific preference. I, I do, you know, I I don't not suggesting I I have a strong position on it, but it just seemed like.

903 "" (0) 05:52:07.290 --> 05:52:27.290 Process ones that it would be part of that bucket. So with, I think there's probably two things that make it different from that. So Tim with robots, question of exclusions, the, the public benefit thing was much broader than, than this. This is simply talking about making this.

904 "" (0) 05:52:27.290 --> 05:52:32.100 Snippets accessible to the person who's consuming them.

905 "" (0) 05:52:32.100 --> 05:52:49.560 And, the, the other aspect of this one is because I think this is such a fundamental part of the application failing to address it and relying on the 3.2, maybe you just ignore the preference thing, probably doesn't fly because.

906 "" (0) 05:52:49.560 --> 05:53:04.620 What you're looking at here is, not just it's a package deal and if you decide to ignore the entire thing, then you're ignoring the entire thing. And that's, I mean, maybe we can.

907 "" (0) 05:53:04.620 --> 05:53:21.210 Carve out space for selective ignoring of the preference, but I, I I think this because this is such a common thing, it maybe we need to address this straight on. Yeah, I mean I I just had I don't know if it'd be common.

908 "" (0) 05:53:21.210 --> 05:53:41.210 Means that we have to address it. It might actually mean that we don't address it because it's going to happen and and has been happening and NO one has ever, you know, when, when, when content is collected 1st, you know, for search applications, there isn't an explicit acknowledgement or you know permission given that your course you can translate the page and make it accessible, it's just it's assumed.

909 "" (0) 05:53:41.210 --> 05:53:57.330 So do we need to now, you know, kind of make these exceptions for these preferences or can we just not, can we not assume that that will continue? And, and I do think that, that there will be partial just you know.

910 "" (0) 05:53:57.330 --> 05:54:16.410 Respectful preferences because once you have content, it's not a question. It's, we're not talking about access, we're talking about use. So some of the uses may go over, some may go under I think there's gonna be a slightly scale of, of, yeah. So I think I think if we, if we include the word verbatim.

911 "" (0) 05:54:16.410 --> 05:54:33.150 Then we basically precluded this very clearly like the word verbatim makes it very crisp and clear that the expectation is that this is not in scope. We leave it out, then maybe there's some ambiguity about that?

912 "" (0) 05:54:33.150 --> 05:54:49.890 But, I'd like to hear from others. Robots next, right? Or next? Yes. Okay put my hand up. My name is Timore Robot. So I think two things. One.

913 "" (0) 05:54:49.890 --> 05:55:05.160 I think that failing to carve out these important uses public interests or accessibility will have a chilling effect.

914 "" (0) 05:55:05.160 --> 05:55:24.420 And will, and so I think it's I think it's important to somehow, whether it's per category or at a top level. 2nd thing is I think that like presumably we want entities to respect these preferences.

915 "" (0) 05:55:24.420 --> 05:55:42.150 And one of the kind of functions of normative preferences like this is, they're, they're paid for in, in in some part by reputational benefit that a.

916 "" (0) 05:55:42.150 --> 05:56:02.150 Someone who's ingesting data, like they, they may wanna say we respect these preferences and and and make that claim. And if we are requiring people to like opt out, ignore, disrespect the preferences for basic accessibility.

917 "" (0) 05:56:02.150 --> 05:56:18.180 That, that makes that incentive, that kind of that social incentive much more difficult, and I think weekends the entire proposal. Thank you.

918 "" (0) 05:56:22.290 --> 05:56:40.590 So, so Martin and I refreshed the PR. Do you want to walk people through the changes you made? Yeah, so, Layer and I were were just talking about making sure that the the conditions are presented in a list. So that's, that's just a simple simple editorial.

919 "" (0) 05:56:40.590 --> 05:56:56.730 Construct and it also addresses, I think the sort of difficulty of the very, very last line of the section, what it refers to because the word conditions is, it just makes it a little easier to follow.

920 "" (0) 05:56:56.730 --> 05:57:11.760 The other thing is this bit that we're talking about just now, which is the the bracket of text underneath which which covers the non substantive changes to the presentation.

921 "" (0) 05:57:11.760 --> 05:57:31.760 Kind of I'm still working on that, but I just wanted to get something in there that we could sort of, Yeah. Thanks. I can agree, I just think the discussion of these here is specific to making search usable.

922 "" (0) 05:57:31.760 --> 05:57:51.570 For the public. It is specifically to help people find things perhaps not in their language of origin or to make the results accessible. Like it isn't for all uses of this content, which is why I think it makes sense to have them here in search. And at least from my understanding that they're built into part of.

923 "" (0) 05:57:51.570 --> 05:58:11.570 Like that opting out of them might be complicated for the way search works. Evan, do you.

924 "" (0) 05:58:11.570 --> 05:58:28.050 Come off mute like after Chris to say what you said to the search. We'll just let go Chris go 1st and then I was just noticing that we're looking at a search definition, but I wonder if anything about the.

925 "" (0) 05:58:28.050 --> 05:58:45.150 About these parts, let's call them is specific to search. Because is the question of transcription, translation, text to speech, another accessibility type features specific to search or is it a more general question?

926 "" (0) 05:58:45.150 --> 05:59:02.190 Thank you. Chris, so I'm gonna focus on text to speech, and this, this has potential uses which are.

927 "" (0) 05:59:02.190 --> 05:59:22.190 Perhaps not necessarily socially beneficial. E.g., you could use a text to speech engine that takes a well known person's voice and puts words in their mouth to have them say things that they didn't say. And, you know, there are news articles where, you know, people this is being used sort of perhaps for fraudulent.

928 "" (0) 05:59:22.190 --> 05:59:26.130 Purposes and, you know, things that.

929 "" (0) 05:59:26.130 --> 05:59:42.480 Really we we are quite harmful to, to, to, to people. So my, my concern with having a carve out within the search category for, for this particular application is that somebody who's building.

930 "" (0) 05:59:42.480 --> 06:00:02.130 A text to speech service that's AI based, uses the A that uses the search category to say, well, actually the thing that we're building fits into this definition and therefore none of the you know the therefore none of the sort of constraints around AI usage apply to us.

931 "" (0) 06:00:02.130 --> 06:00:21.480 So I think we have to think, you know, how this might be used in, you know, by people who are kind of look, you know, looking for gaps in the the definitions that we provide in order to.

932 "" (0) 06:00:21.480 --> 06:00:38.970 You know, to, to, to find ways to, you know, to avoid sort of respecting the, you know, the preferences, so if there's some way to do that, I, so that's, that's kind of one of my objections towards sort of including.

933 "" (0) 06:00:38.970 --> 06:00:58.970 You know that in particular, whether, you know, if there's a way to say, you know, well, these these things are, you know, bound to it being a search application and not some other kind of application, then, you know, perhaps that alleviates the concern some.

934 "" (0) 06:00:58.970 --> 06:01:16.050 More, but I just feel like, you know, including these kind of carve outs, you know, opens the door to sort of unwanted, you know, applications, not unwanted or you know as sort of undesirable, you know, consequences.

935 "" (0) 06:01:20.550 --> 06:01:39.390 You? Just to say I I I thanks Chris, and I I agree that these are real societal problems and issues of deep fix and, you know, false content and everything. I don't necessarily think that this is a meaningful or necessary place to establish controls for that.

936 "" (0) 06:01:39.390 --> 06:01:58.290 I'm not sure even like the publisher of a website that has a picture of somebody should be able to say, you can't use text to speech to put words in the mouth of this person or the publisher of the website that has the text on it says you can't make speech, like it just it seems like a very ill fit for a very serious problem.

937 "" (0) 06:01:58.290 --> 06:02:15.990 And I I I think it probably does more harm to our project than good to have to take on the mantle of, you know, meaningfully addressing those issues and I'd urge us to not to concentrate on that. Justin.

938 "" (0) 06:02:15.990 --> 06:02:33.330 Yeah, I, I can appreciate that point of view on fix, I don't think this is the forum where we're gonna fix that and people are gonna do that. They're gonna do that regardless of the preference, they're gonna ignore it as the preferences allowed. I think that's for a matter of law, but with the purposes of this text transcription translation or text to speech.

939 "" (0) 06:02:33.330 --> 06:02:52.560 For the purposes of accessibility would then provide the intent, you know, something new to that effect. I think I think it's very important that we protect the use of accessibility features as a matter of saying, you know, the preference specifically doesn't stop you from doing that cause I think that's something we need to.

940 "" (0) 06:02:52.560 --> 06:03:12.180 Protect where we can. Next.

941 "" (0) 06:03:12.180 --> 06:03:28.500 Paul? Yeah, I think I want to make the same point as many other people wanted to make, basically Meredith's point that sort of like if we, if we assume like complete bad face actors as.

942 "" (0) 06:03:28.500 --> 06:03:45.030 If we need to con, if we assume that our preferences need to be like to sort of condition the behavior of hectors, like we'll never get people. The other thing is there is a class of things here that require completely different approaches.

943 "" (0) 06:03:45.030 --> 06:04:05.030 European approach to sort of stuff is legislation, like I don't think we particularly need this like that maybe different in other parts of the world, but that is also outside of our charger, right? Like we had that before with the notify or notify apps or something like there's just things like that we are not here.

944 "" (0) 06:04:05.030 --> 06:04:08.340 Harder to address and sort of bringing them in.

945 "" (0) 06:04:08.340 --> 06:04:28.340 That like doesn't help us in finding workable definitions for the things that we are charged to address. So I was looking at the definition that we have here and I was thinking as to what was troubling me about this and I think like kind of have a sense of, right? Although I'm.

946 "" (0) 06:04:28.340 --> 06:04:48.240 Not sure exactly which sentence here triggers that. Right now in the search world, there's a whole bunch of behavior and preferences that can be expressed that can be quite restrictive. Like e.g., there is something like NO snippet that people can express or a max snippet that people can express. Image sizes can be expressed.

947 "" (0) 06:04:48.240 --> 06:05:06.540 How do we reconcile this language with the existing controls out there? Because this seems to alter how search behavior happens.

948 "" (0) 06:05:06.540 --> 06:05:26.490 Give me something I can use. I just gave you something you can use. This definition, e.g., might override other things where people have said things like NO snippet or max snippet equals 30 characters or many of the other controls that are out there that are quite heavily used.

949 "" (0) 06:05:26.490 --> 06:05:43.380 But you can also invert that if if without the words. No, it doesn't. Also, it says changes to those things. What changes? Read the text, non substantive changes.

950 "" (0) 06:05:43.380 --> 06:06:02.670 So those things, if the snippet doesn't exist, then it can't be changed. If a snippet does exist, it's one of the options. Sure, but if if you have max snippet that says, you, you or NO snippet or whatever, what have you, there's gonna be NO snippet to translate or.

951 "" (0) 06:06:02.670 --> 06:06:22.670 Presents in a different way. But if there is a snippet there, whatever controls you have over the snippet that already exists will then flow through to the changes that are presented. So if if text to speech is used, it will it will only be able to read out the things that are present. I don't know I understand your point, point me to the text where it says.

952 "" (0) 06:06:22.670 --> 06:06:41.460 Changes. The. What changes? Changes is not defined here. It's in the dictionary. Oh, really? Yes. I don't think you need a special definition for that word.

953 "" (0) 06:06:41.460 --> 06:06:58.110 Disagree because this is all about changing current behavior. We are trying to redefine what search does here. Tell tell me what is wrong with this, with the statement, non substantive changes to the presentation, titles, excepts.

954 "" (0) 06:06:58.110 --> 06:07:16.980 All snippets from assets is included in in this definition. So in the current regime, when you specify something like a max snippet, you don't have to wonder about what is non substantive about it. This term non substantive is a.

955 "" (0) 06:07:16.980 --> 06:07:35.610 Not defined term here. What is non substantive? It's defined on next sentence. Oh, so, this includes ensuring that users are able to unders understand the material. I put this together in 5 min. So, it's ok.

956 "" (0) 06:07:35.610 --> 06:07:55.610 So if you upload this a little bit Krishna, right? So if you, let's say somebody says snippet equal to 40 words, right? Yeah. And, so as long as the transformations on the 40 words are the stuff listed there, translation, text to speech, whatever, how does it contradict this really the question? So if you can kind of talk about like how.

957 "" (0) 06:07:55.610 --> 06:08:01.830 So this stuff is contradictory to something somebody said in the search controls.

958 "" (0) 06:08:01.830 --> 06:08:21.830 Which should take precedence like they combine, right? Like so I think the idea is like you apply the search controls and then you like you can do some transformations on the text that pass the search controls how I read the text, right? Like, so if you think that's not how you read it, I think we can change the thing, but at least that's how I read it. I think that's kind of.

959 "" (0) 06:08:21.830 --> 06:08:41.160 My problem, right? Like, so in the other definitions that, that we had, there was some level of specificity about what would constitute that kind of a substantive change. But here it's like kind of yeah you you have to reconcile it in your head. So I think that that.

960 "" (0) 06:08:41.160 --> 06:08:57.330 There's a general issue here of the relationship of this control to pre existing controls, whether it's, you know, they enhance it or this contradicts them or or whatever, we probably need to at least consider talking about the relationship is what you're saying.

961 "" (0) 06:08:57.330 --> 06:09:14.490 For that matter, there is another control called data NO snippet. You can use to achieve the exact same thing. So how do you reconcile all of this? We can also say it's not the function to reconcile things over here, that's fine.

962 "" (0) 06:09:14.490 --> 06:09:34.490 But people are using these things today, yeah. I, I think it's it's useful to braise the issue and then perhaps do an exploration of what the interactions might be before we kind of have a bigger discussion about it. Because I can imagine that, you know, part of the answer is gonna be about specificity, e.g., if that, you know, NO snippet is attached to a very specific piece of data and you discover the.

963 "" (0) 06:09:34.490 --> 06:09:41.400 You know, the search control through a much more general mechanism, we want to think about the implications of that.

964 "" (0) 06:09:41.400 --> 06:09:56.850 Just bringing it up because it's like bothering me. Could you go ahead and create an issue for that perhaps? And then that can be the home for that discussion. I can attempt to address that. Yeah, I do think we probably need to do some discovery on it 1st time.

965 "" (0) 06:09:56.850 --> 06:10:12.060 You're gonna try. It's just one sentence Timid Ruba? Timid Robo.

966 "" (0) 06:10:12.060 --> 06:10:29.280 Listening to this conversation, it, for me personally, maybe it's just problems with my brain, it would be clearer if in 2nd sentence, this was replaced with non substantive changes, just so that it's very clear out of the 1st sentence.

967 "" (0) 06:10:29.280 --> 06:10:49.280 What this is referring to, you know, there's NO interpretations. Thank you. Yeah. Thank you. Yeah? I just test it in the chat. I think part of the struggle with the block of text. Sorry, I think part of the struggle is this.

968 "" (0) 06:10:49.280 --> 06:11:07.830 Like block of text, and I wonder if it wouldn't be easier to start with Search is X search allows AEC. Search does not allow and so I I've added a comment on that issue.

969 "" (0) 06:11:07.830 --> 06:11:23.820 It tries to restructure it and hopefully makes it clear what is included in the category isn't because I think just trying to read the paragraph like it talks about what's a condition, it talks about what's not included, it talks about non subsident changes, talks about AI training and they're all sorted.

970 "" (0) 06:11:23.820 --> 06:11:42.300 Not clearly related Christmas at an old hand?

971 "" (0) 06:11:42.300 --> 06:12:02.300 Oh, it's, it's my hand raised up from before. I need a AI feature for that. Nate, I'm just a small thing. Yesterday, one of the proposals had a title of ungenerated search, and I that's 1909.

972 "" (0) 06:12:02.300 --> 06:12:18.840 1909. Yeah. And so, and so maybe, maybe we consider that here at the title just something because there's also AI search. I decided against that.

973 "" (0) 06:12:18.840 --> 06:12:34.740 For various reasons. I think it's just easier to not have the extra verbiage around it. Why didn't have you seen the comment on the.

974 "" (0) 06:12:34.740 --> 06:12:54.740 You are from Leo. Let me do so many things, give me You are there let's see.

975 "" (0) 06:12:54.740 --> 06:13:13.350 Was there coming up? Oh yes, ok.

976 "" (0) 06:13:49.470 --> 06:14:09.470 Okay, so, I mentioned this to Layer as well. The 3rd point here is not a condition.

977 "" (0) 06:14:09.470 --> 06:14:25.590 It's simply a statement, and so that doesn't belong in the list. But it's it just says permissible implications. So NO it's it's basically, Yeah, I think permissible is.

978 "" (0) 06:14:25.590 --> 06:14:40.980 Mistake but and I don't know that it uses that word, yeah, so.

979 "" (0) 06:14:40.980 --> 06:15:00.090 The reason that I want that out is the text that I have right now is, if this use is allowed, that's the structure of a sentence as it stands. If this use is allowed, then these things are possible.

980 "" (0) 06:15:00.090 --> 06:15:19.050 Which is not a good picture. The restrictions on the scope of the thing are the 1st two, and that, that sentence or that paragraph or that last bullet there needs to refer to the the other two things as being restrictions that apply on the on the training of models.

981 "" (0) 06:15:19.050 --> 06:15:37.140 So that's important. I'm not sure about I'm not sure why you didn't include the links in that list.

982 "" (0) 06:15:37.140 --> 06:15:56.490 Yeah, so the way that I've structured it in the current one has two restrictions. One is that you only use excepts and snippets.

983 "" (0) 06:15:56.490 --> 06:16:13.950 Whatever that means, and that you provide links to the original location where it was taken. And those are the, the, the relevant restrictions. And so, I think, can I think I prefer that.

984 "" (0) 06:16:13.950 --> 06:16:29.730 Think about the last one, I'm gonna hit edit so that it refreshes the content. Sure. But I don't think this addresses the the main concern that I had, which was that, that, yeah.

985 "" (0) 06:16:29.730 --> 06:16:45.120 So that, that bottom paragraph, the one that says prefer preference to allow this category of use, it says provided that the asset and the outputs of any processing is exclusively used by a search application according to the other restrictions of this section.

986 "" (0) 06:16:45.120 --> 06:17:03.330 The current text now, says to the above conditions and and that refers to the items that I have in the list, but your conditions are spread now. Yeah, I'm not I'm just trying to break it out a little bit to address your concern.

987 "" (0) 06:17:03.330 --> 06:17:20.310 I, I think it would be good if we focus on kind of substantive issues here. That's editorial. Where did it go? Did I pass it? Can you pass it.

988 "" (0) 06:17:20.310 --> 06:17:40.110 So I put in a. Try to get to Christian's point there. Are way way down. There it is. We'll ways. It's like it's it's sipping the bits through a straw.

989 "" (0) 06:17:40.110 --> 06:17:57.930 So the transformation thing gets pretty, pretty narly out of this, but the sentence that I added there was where existing controls restrict the presentation of these items such as limitations on snippet size does apply before any transformation.

990 "" (0) 06:17:57.930 --> 06:18:14.040 Which I think addresses the concern that Krishna raised. Yeah at least there's a resolution there, yeah.

991 "" (0) 06:18:14.040 --> 06:18:29.970 And I I I use the word accessibility in the 1st sentence as an, as a, as another change there, which I think gets why these transformations exist.

992 "" (0) 06:18:29.970 --> 06:18:49.970 Yeah, is included the right word? Second sentence. Well, I I want to avoid using allowed in the definitions of these things because of the way in which the rest of the architecture works. But included is a little bit ambiguous.

993 "" (0) 06:18:49.970 --> 06:19:05.910 Yeah, can you use colored? Now we're editorializing and I don't think that's really a huge.

994 "" (0) 06:19:05.910 --> 06:19:21.060 Perhaps you could refer back to applied, but anyway, yeah. So we still have the verbatim, well that would come out of this text goes in. Okay.

995 "" (0) 06:19:21.060 --> 06:19:39.750 There's three options that I see here. One is that we put verbatim in and remove that larger paragraph, this addendum, another one is that we remove that paragraph and remove the word verbatim.

996 "" (0) 06:19:39.750 --> 06:19:59.750 And leave some ambiguity about what it means to have exports into this and the final choices to take verbatim out and take the paragraph and, and use that paragraph for something like it. I don't like the ambiguity option, but it's a possibility as well.

997 "" (0) 06:19:59.750 --> 06:20:04.710 I ask what work snippets is doing that isn't doing?

998 "" (0) 06:20:04.710 --> 06:20:20.370 Editorializing? Sure. So, one of the things I think we're trying to sort of group our way towards is.

999 "" (0) 06:20:20.370 --> 06:20:39.630 If users knew the verbatim words to search for to find the information that they would need for everything, they would use them. There are a lot of cases in which users do not know their verbatim words to search for to get the answers they need. So there's this whole bundle of things that help.

1000 "" (0) 06:20:39.630 --> 06:20:59.070 That gap. Like when you search for something and you use like your search for a quote and you're using the wrong words, the thing you're searching for still shows up. That was I think someone put an example in the chat. So trying to caption this to like just which gap filling technology we're talking about is a.

1001 "" (0) 06:20:59.070 --> 06:21:16.740 I think an over focus on the technologies. But I, I can't, I need to like do anything. I think I think verbatim applies to the content that's coming from the site, not to the search itself. Isn't the match between those two part of what the I don't think so. No.

1002 "" (0) 06:21:16.740 --> 06:21:33.000 So in your example where you search for the, the wrong phrasing of particular quote, then you're looking for that text on the internet somewhere, you get the quote wrong in your search, but the quote is correct and the snippet that's presented, and that would be fine.

1003 "" (0) 06:21:39.270 --> 06:21:55.920 I put an example in the chat where I was trying to find, actually it turned up from the millers table.

1004 "" (0) 06:21:55.920 --> 06:22:10.950 And I didn't really know what I was searching for because the language was really old at obscure. The only reason I recognized it is because the snippet wasn't the original old language. It wasn't wasn't updating of it.

1005 "" (0) 06:22:10.950 --> 06:22:26.160 I don't really think that that's an accessibility or a translation. It was, there was a snippet that was actually useful as well. Yeah, I think that falls under the translation course in this case. That language is so alien that you've gotten a translation.

1006 "" (0) 06:22:26.160 --> 06:22:44.220 Okay, I think your question was more to input versus output where someone puts something into a search and it's ambiguous to the output, but if we allow non generative search is indexing whatever the methodology maybe.

1007 "" (0) 06:22:44.220 --> 06:23:03.090 This is specifically speaking, this is speaking specifically to the output of that instead of the word verbatim and again I know we're not trying to editorialize, you could remove verbatim and and change snippets from the original source asset. I think without the word verbatim.

1008 "" (0) 06:23:03.090 --> 06:23:19.500 Where are these search and snippets coming from? They could be generative unless we specify otherwise. So you could say from the original source asset or you know, but I think we need something there.

1009 "" (0) 06:23:19.500 --> 06:23:36.870 I just wanna relate to earlier comments, I want to caution us against trying to evaluate whether we think the search products defined by our definition is more or less helpful.

1010 "" (0) 06:23:36.870 --> 06:23:53.040 To an end user and kind of a generic way that like an AI powered search engine or fully AI powered search eng because I don't think that's what is under consideration here. It's known that this restricted definition comes with trade offs.

1011 "" (0) 06:23:53.040 --> 06:24:13.040 And so, there's going to be times when we are excluding things that people find helpful, but that's a known trade off.

1012 "" (0) 06:24:13.040 --> 06:24:21.840 Yeah exhausted the queue? Yeah.

1013 "" (0) 06:24:21.840 --> 06:24:41.840 So if I understand it, Martin, your suggestion, your your 1st suggestion here is to remove verbatim and include this paragraph, so that would be, that would be an option. Yes. I'm not trying to prejudice anything here, I'm just putting the, the options out on the table. I think the other option is to.

1014 "" (0) 06:24:41.840 --> 06:25:00.030 To put the word verbatim in and not have that remove the paragraph, right? And the 3rd is to not have anything and leave it. And the the other option is the ambiguity option. I don't know that that's a good one, but I I think, yeah, we we've gone down this road of fair wastes because there are concerns about the ambiguity. Exactly.

1015 "" (0) 06:25:00.030 --> 06:25:20.030 Should we hazard another poll? I think it's either verbatim and remove the paragraph or remove verbatim and include the paragraph.

1016 "" (0) 06:25:20.030 --> 06:25:39.570 Or your preference. Or something else, I guess. Sorry? We're both or another. Well both would be problematic. So.

1017 "" (0) 06:25:39.570 --> 06:25:56.100 We can or we can just ask for objections of doing one, like just would you prefer like taking out verbatim and like we can pull on it, just like ask for objections on it, makes it easier. I think, I think the easiest poll would be.

1018 "" (0) 06:25:56.100 --> 06:26:11.610 Verbatim or not. Well, that'll give us some information. Certainly. Yeah. Okay. And the not verbatim plus the paragraph or is that a separate one? And and then we can talk about whether the paragraph inspect. Right.

1019 "" (0) 06:26:11.610 --> 06:26:27.420 So we'll Paul on verbatim or not verbatim? Not verbatim implies that we we may have caveats or or explanation like this.

1020 "" (0) 06:26:27.420 --> 06:26:46.860 Or or like something else. I don't think we'll have a NO preference option because if you don't know, just don't vote in this one. Warren you look confused? Is that just the normal? Yeah, I usually look confused, but I try to figure out how something fits into Martin and others definition.

1021 "" (0) 06:26:46.860 --> 06:27:04.590 So I was trying to find like there's a well known speech, say Christmas Day Shakescare. Couldn't remember what it was called, but I searched for famous speech outnumbered borrow. And the text that I'm searching for doesn't show up anywhere in the speech.

1022 "" (0) 06:27:04.590 --> 06:27:20.100 The snippets don't really have it because it's not in the original text. But something told me that it's St. Henry the 5th st. Prisman Day speech. That's not a generated from a summary. It's not a snippet.

1023 "" (0) 06:27:20.100 --> 06:27:40.100 But it got me to actually be able to find what it was that I was looking for. So that is AI generated. It's not using a snippet, but it helped me figure out how to get to where I want to be a bunch of other links because it inferred from what I was trying to search for.

1024 "" (0) 06:27:40.100 --> 06:27:59.820 Use the eye on your input, and generated output that told me what to get to, right? And that I Would it was a useful thing because now I actually found the point I was looking for. It generated a bunch of texts so I knew what I was looking for. What was the text it generated?

1025 "" (0) 06:27:59.820 --> 06:28:15.240 The most famous speech regarding being upnumbered is King Henry's King Henry the 5th state Chrisbane Day speech, which refer refrains blah blah blah, and then it had linked to Wikipedia and to a poetry sake that had it. Do you know the great thing about that?

1026 "" (0) 06:28:15.240 --> 06:28:35.240 That speech is in the public domain. Yeah, yeah, yeah. But but ok, I agree fair enough. But what I searched for wasn't actually a snippet or something, but yeah, you're right that Well it's in the public domain that doesn't mean that people who put it somewhere wouldn't have tagged it as you can't search this. Correct. Right. Right.

1027 "" (0) 06:28:35.240 --> 06:28:39.720 Could be placed somewhere where I still don't have access to it, even though it's in the public domain.

1028 "" (0) 06:28:39.720 --> 06:28:59.250 But there's nothing in the thing that actually helped me find it. Calls life. Goal's life. If you don't know or don't care, don't vote. So I think the answer is that would be, and that is a very useful feature and it is beyond the scope of this definition we're talking about.

1029 "" (0) 06:28:59.250 --> 06:29:16.980 Well, NO, but it's if I can't, if I can't use the text, I don't think it is beyond that. Yes, yes, it wouldn't be allowed by this and I think that that's an issue. Right. I think that's a major failure because.

1030 "" (0) 06:29:16.980 --> 06:29:36.980 Isn't another site it came from? It didn't come from anywhere. Right? It's. This is knowledge that's baked into the model. This know, this is knowledge that's baked into the model. It's not knowledge that was generated. It had information in the model because it learnt that outnumbered is the sort of point that this whole model.

1031 "" (0) 06:29:36.980 --> 06:29:42.420 But it is because what I searched for was you said it's not relevant.

1032 "" (0) 06:29:42.420 --> 06:30:02.420 Famous speech about being upnumbered. This is the exact speech. But that's model upput. Put it? That's marterload output. It didn't come from somewhere else. There's NO preference attached. Okay, Christopher? Just a question, so removing, are we we're keeping the non substantive transformations? I think that's the next step, but once you like but I think that.

1033 "" (0) 06:30:02.420 --> 06:30:14.850 Should be like a natural follow up for the removing because it doesn't make sense if you.

1034 "" (0) 06:30:14.850 --> 06:30:34.850 Okay, I think I would say a significant preference for removing the word. Yeah, so right now we're 78 % to removing verbatim. Those 22 % of folks and and and there's also a substantial number of folks who haven't voted so if if you want to please do so soon, but I'm also.

1035 "" (0) 06:30:34.850 --> 06:30:43.830 Interpreting that as a lot of people don't have a preference or or don't have enough information. Those 22 % who want to keep verbatim in if if we could hear.

1036 "" (0) 06:30:43.830 --> 06:30:59.190 Why you think that's necessary, and any thoughts around that, that would be really helpful if anybody wants to step up. Warren and then Justin? Oh, sorry, but Justin? I I had previously stated if we remove verbatim then.

1037 "" (0) 06:30:59.190 --> 06:31:17.970 That we it doesn't say what part of the source has it could be generated we could fix we can fix that. Yeah like if it said original, as I was saying before, then I would be ok. Yeah. The removing it doesn't preclude other fix UPS that's editorial.

1038 "" (0) 06:31:17.970 --> 06:31:37.970 Anyone? Going once five. Well, let's let's what were the finals? 75 % 708I think it was. 7525 7525. Okay, thank you.

1039 "" (0) 06:31:37.970 --> 06:31:40.290 Out of 28, like 21 said.

1040 "" (0) 06:31:40.290 --> 06:31:59.850 Remove 27 set and don't remove. So again, this is not we're not looking for consensus yet, we're looking for what direction do we develop this proposal in that we can eventually try and get consensus on. So, for now let's remove verbatim, then I guess, you know, let's look at this.

1041 "" (0) 06:31:59.850 --> 06:32:17.670 Paragraph, and see if that addresses if that's something that we're we're, we're happy with. So now we're looking at this one.

1042 "" (0) 06:32:17.670 --> 06:32:34.620 Christopher? I was just gonna say like Brad's suggestion for alteration transformation. So I addressed that in the chat. I think that transformation allows us to engage with.

1043 "" (0) 06:32:34.620 --> 06:32:52.980 Christmas concern much more directly. If you if you imagine alteration, the, that means a change other than that potentially means a change other than than that necessary for.

1044 "" (0) 06:32:52.980 --> 06:33:08.430 The accessibility purpose that you're trying to, to achieve. So transformation like trans like a translation is take from one take the same meaning from one language and put it in another one. Whereas alteration means if it says one thing.

1045 "" (0) 06:33:08.430 --> 06:33:25.380 An alteration includes negating the meaning of it, which is not at all what we're looking for. Okay, I guess in the translation would make more it is just didn't seem to make sense cause non substantive and completely changing. Yeah, but I get what you mean.

1046 "" (0) 06:33:25.380 --> 06:33:45.380 You can transform something into something completely different too. I don't think I mean I I I think It's we use the word change? Yes. I I I don't I mean I think I don't think it's less, I don't I don't think it raises differentiation, but I think transformation.

1047 "" (0) 06:33:45.380 --> 06:33:48.900 Starts to sound a little fairy and I.

1048 "" (0) 06:33:48.900 --> 06:34:08.900 Sure think that people are gonna, you know, sort of like load it up with some, some of their context and I don't know if that's helpful. So, I mean what we're looking for at this point is substantive kind of, you know, reservations or emissions in this text. I think the assumption going forward is that, you know, the editors will take it and refine it, make.

1049 "" (0) 06:34:08.900 --> 06:34:25.380 Get communicate well related to the rest of the text properly, people will have a chance to look at it, consider raise issues, make suggestions, but what we're looking for now is, is do, do we see deep and troubling.

1050 "" (0) 06:34:25.380 --> 06:34:44.910 You know, problems or or or emissions in in this text And again, i'll refresh and go through the awkward dance that get up next we go through. Yeah, go through that dance again. I have made it slightly shorter.

1051 "" (0) 06:34:44.910 --> 06:35:00.090 I'm done. It's not. There it goes.

1052 "" (0) 06:35:00.090 --> 06:35:19.920 Readly are we losing you?

1053 "" (0) 06:35:19.920 --> 06:35:39.120 Yeah, but I'll be I'll be remote tomorrow. Be remote tomorrow? Okay, be remote, yeah. Thank you. Yeah.

1054 "" (0) 06:35:39.120 --> 06:35:56.400 Is there a poll? Not at this point. I think would you like to put one up? Well, I think we can take the paragraph or not, but I think really is what we're up, up to. Is there any other concerns coming up? Yeah I I've tempted just to ask for objections, but.

1055 "" (0) 06:35:56.400 --> 06:36:13.620 If people want to get the information of how much support there is in the room for it, sure. Yeah. So Maybe add the paragraph or something else. The paragraph doesn't provide enough info, but I can.

1056 "" (0) 06:36:13.620 --> 06:36:33.330 Oh, don't copy that. Non substantive changes paragraph. The selected paragraph on the screen, but I really wish my new glasses had arrived before this meeting.

1057 "" (0) 06:36:34.560 --> 06:36:54.150 They're really cool glasses. No, but they have a stronger prescription than they are really cool. Like being able to see is really cool. Yeah. I highly recommend it. I just don't have a lot of experience with it recently. Yeah. Like being able to see I'm waiting for my extras insurance to re up in July.

1058 "" (0) 06:36:54.150 --> 06:37:11.820 I was, I must be getting old. I went to the ophalmetrist and like, so do I need new script and he's like nah, your eyes are exactly the same as last time. So you you get past that pretty soon. Welcome.

1059 "" (0) 06:37:11.820 --> 06:37:31.230 Yeah Cherish these years Mark. Yeah Paul is online. Old enough. Oh. Just to clarify this.

1060 "" (0) 06:37:31.230 --> 06:37:51.230 Include the whole thing or nothing. This is well really it's inclu This is good. Let's con go forth with this or I have serious reservations or, you know, I prefer to take a completely different approach. If you think you can fix it, then you can say this. Yeah, if you don't think it's fixable. Not that it is pristine and perfect and and should never change. That's not what we're trying to do.

1061 "" (0) 06:37:51.230 --> 06:38:03.990 Substantive changes that make. If you have something completely different substantive changes.

1062 "" (0) 06:38:03.990 --> 06:38:23.010 And Justin, if it means that you're gonna come up with a shorter replacement for it, yeah, please go ahead. I asked Lila too so like you do getter that colors the same thing, please make this well we can make this shorter, you know, in the process going forward. I would not, I would hope that we're not, you know.

1063 "" (0) 06:38:23.010 --> 06:38:42.450 At that point. If we need to pull for every editorial change, we're in a pretty bad place. Yep. So if you haven't pulled yet, please do so so soon.

1064 "" (0) 06:38:42.450 --> 06:38:57.810 Right now we're at 83 or 3 % for yes and 17 for now. Does anybody who said NO want to say why? Or what they would suggest instead? I still just don't think that most of that should be at the search level.

1065 "" (0) 06:38:57.810 --> 06:39:17.810 I Wondering if it's sort of like nothing in the standard to be contributed to pinge on people's ability to do those things kind of framing that is not specific to search would be interested. I would in that case vote to include it and then factor it out.

1066 "" (0) 06:39:17.810 --> 06:39:30.540 Not when we included in every other session. If, if we can get to a point where we have comfort that, you know, it's not gonna be, I think part of the reasoning here is, is this is a very specific application and we want to be crisp about it. But if we get that crispness.

1067 "" (0) 06:39:30.540 --> 06:39:50.540 Through other mechanisms, it can of course be refactored out. Yeah, I don't know if that's anything that's fine. Anybody else? I think everybody else said. Oh pardon me, I totally agree with what Kevin said we put this text in that is the way to do it.

1068 "" (0) 06:39:50.540 --> 06:40:10.230 It applies to everything, right? Why don't I still hang up on Shakespear? I know Cause I think that there are useful transformations that could happen, which don't strictly fall into, you know, for the purpose of accessibility.

1069 "" (0) 06:40:10.230 --> 06:40:25.829 Like in terms of accessibility and understandable understandability, but I don't know of text that I can provide that would do that. Can you go away and think about it? Maybe raise an issue?

1070 "" (0) 06:40:25.829 --> 06:40:44.429 That's why we make our progress here. Yeah. Because cause, you know, at this point we're just, we want to come out of this meeting with sure. Okay. So I don't have the right text also I understand how to motivate. Also because I understand like I know what we're trying to accomplish, I just don't think that this text doesn't and.

1071 "" (0) 06:40:44.429 --> 06:41:04.429 I mean, it's not quite the right I think part part of the purpose of this text is just so that when we publish something, you know, another draft out of this meeting, people isolate reading it in isolation can understand what the intent is. Yeah. Thank you. Andrew. So I'm one of the nose I I'm not gonna stand on it, but I I am deeply concerned about the substantive not.

1072 "" (0) 06:41:04.429 --> 06:41:10.589 Substantive distinction I don't understand it. I don't know how you get there and I think that just about everybody.

1073 "" (0) 06:41:10.589 --> 06:41:27.299 Is going to come up or just about anybody who objects to this is gonna come up with some corner case where they can show that a non substantive thing is substantive for some reason. Your examples that's used here and there are lots of places where if you translate something. Oh, yeah.

1074 "" (0) 06:41:27.299 --> 06:41:44.879 Yeah, so if we, if we struck that work, would that address your concern? I mean, like I said, I'm not gonna stand my interest that I'm growthy at this point, so. Oh, welcome. Can we get you a cookie and maybe that'll help? It sounds like it would help Lauren's concern a little bit as well. I don't think there's any need.

1075 "" (0) 06:41:44.879 --> 06:42:04.879 To keep that word given the other context that's been put in in there just I think I think there's an attempt to draw a distinction that is actually problematic in here, but I think so I I'm not gonna yeah yeah I get that. And again, I think, you know, if, if you can.

1076 "" (0) 06:42:04.879 --> 06:42:13.049 And go away and think about that and articulate that in an issue, you know, we can come to that. I'll try ok thanks.

1077 "" (0) 06:42:13.049 --> 06:42:33.049 I mean I like non substantive cause I think it's important to to help clarify that what we mean by that. I really debated which way to go on this one. I think that these are important things that that we should care about. I kind of tend to agree with Kevin. I think I've been somewhat persuaded on that and that that was part of the motivation. The other thing.

1078 "" (0) 06:42:33.049 --> 06:42:34.529 I am still concerned about.

1079 "" (0) 06:42:34.529 --> 06:42:54.529 Is the potential for, you know, this being used to find consent for something broader. And I would just say generally as a comment, if we have a grounding category or something that becomes a lot easier to, to feel very confident about like a hundred times easier for me to just be like wait, good because there's this other thing that it's contrasted to.

1080 "" (0) 06:42:54.529 --> 06:43:00.599 So that's what I'll say. I, I think like to go back to what Martin said, right? When we have that, let's kind of.

1081 "" (0) 06:43:00.599 --> 06:43:20.599 Yeah, if you do have a Well we have that when we have the three.two discussion resolved, there's a lot that might change this over time. Any integration of something like a top level category or an inference category or or something like that changes the entire architecture of everything and we're gonna need to go through and very carefully make sure.

1082 "" (0) 06:43:20.599 --> 06:43:40.559 Work that those things fit together properly where we stand right now, this has a particular shape and I'm roughly ok. Okay, and and if it gives people great comfort, we can, we can always create an issue to say come and reconsider this once, you know, other parts of the specification are more fully fleshed out. So.

1083 "" (0) 06:43:40.559 --> 06:44:00.559 Tom, did you wanna speak up? Yeah, I changed my mind because I voted to keep it, but I don't think it should be in there. You need to actually speak up. My voice is a bit screwed I've got a bit of a cough, sorry, but yeah, I did vote to keep it in there, but I.

1084 "" (0) 06:44:00.559 --> 06:44:11.069 I think I've changed my mind because I think it, I don't think it really needs to be in there because if the if the users are legitimate enough, then, you know, to override that verbatim requirement then.

1085 "" (0) 06:44:11.069 --> 06:44:29.369 They don't need a cart now, do they? So can I ask for you and I guess for everyone else, given the discussion we've had over the last several minutes, and the possibility that we might factor it out depending on other changes in the specifics. This is all quite dynamic.

1086 "" (0) 06:44:29.369 --> 06:44:48.299 Is is it just that you don't think it's necessary or you think it's actually causing harm? Well I mean, the, the non substantive thing is, is weird and like it also doesn't define substantive changes, which is like that'll be a categorization headache.

1087 "" (0) 06:44:48.299 --> 06:45:06.029 At some point, if we just if we're talking about non substantive here decide what substantive it is. So perhaps we, we, we need do need an issue about that non substantive word, yeah. Yeah. Or just drop it because.

1088 "" (0) 06:45:06.029 --> 06:45:22.499 You know, they should be kind of taken through like a matter of course, nope? I'm hearing it as a descriptor for folks, so when I felt the word or you want to felt the whole bigger, the whole paragraph I would say.

1089 "" (0) 06:45:22.499 --> 06:45:42.499 One? Yeah, I just pasted in the chat the two different different two different definition of accessibility. One of those is for people who have a disability, and that's how I've been reading our usage of accessibility, right? Like assistant devices, screen readers.

1090 "" (0) 06:45:42.499 --> 06:45:43.289 Etc.

1091 "" (0) 06:45:43.289 --> 06:46:02.999 But what Martin pointed out is there's also the quality of being easily understood or appreciated. So, and I hadn't been viewing that usage and if one does I read it that way, that's much more fine. So a a modern interpretation of the word accessibility is much more inclusive than just people with disabilities for everyone.

1092 "" (0) 06:46:02.999 --> 06:46:22.999 Sure, just what I've been thinking of like making my website accessible, I had been like which is a standard thing or accessibility is primarily the 1st usage. And so I think I at least had been reading it through that. So accessibility folks talk about it.

1093 "" (0) 06:46:22.999 --> 06:46:34.079 It being useful for people, e.g., who have temporary disabilities or people who are under great cognitive load and therefore need a different view of the data that it's a very broad meeting.

1094 "" (0) 06:46:34.079 --> 06:46:54.079 But are you saying the concern is that 2nd definition would allow unexpected use of I'm saying 2nd reading it with 2nd definition addresses my concerns because, you know, making it more understandable or more readable, not, you know, for people who have a disability or challenge or anything.

1095 "" (0) 06:46:54.079 --> 06:46:56.999 This is my concern of like what if we just wanna make it so people.

1096 "" (0) 06:46:56.999 --> 06:47:16.999 You know, Regardless of capabilities. Perhaps just some wording changes rather about the user's ability to access the data or something like that. Yeah, they're not yet it was just time hand ok. So Martin, do you know what to do next on this?

1097 "" (0) 06:47:16.999 --> 06:47:25.109 I'm a bit like little I'm not sure about the non substantive word, like I don't know if he said that I think that.

1098 "" (0) 06:47:25.109 --> 06:47:45.109 I think I've heard that it's useful. I think I've heard that it's harmful and I think both of them are right. And, I, I don't know that I have enough information on whether that word's there or not. I think we can say and that's fine.

1099 "" (0) 06:47:45.109 --> 06:47:54.629 But otherwise, I think I'm just gonna remove the square brackets on the paragraph and perhaps just move the square brackets to the word non substantive.

1100 "" (0) 06:47:54.629 --> 06:48:12.089 Oh or do that, but, yeah. Okay. But I don't think any of those views were particularly strongly held, and I think that I think John had a comment in in the.

1101 "" (0) 06:48:12.089 --> 06:48:28.859 In the chat that sort of suggested that we identify what it is that we mean and sort of expand that to mean those that do not alter the the the meaning or intent of of the content, but I'm not sure that we need to say that explicitly.

1102 "" (0) 06:48:28.859 --> 06:48:44.669 Are are you, do you wanna take another editorial pass at this or is this something that's almost ready for the list? I would say take this to the list at this point. I don't, we're kind of at the point of diminishing returns. Okay. And, and I think, you know.

1103 "" (0) 06:48:44.669 --> 06:49:04.669 The idea being that we're gonna have where we discuss it on list and then hopefully have get to a consensus call. I really hope we don't have to wait till our next interim before we can do that. And I think we'll we'll be very careful about how we establish that consensus. We want to leave the door open to further editorial adjustments and refactorings.

1104 "" (0) 06:49:04.669 --> 06:49:07.979 We want to just have consensus on the meeting.

1105 "" (0) 06:49:07.979 --> 06:49:27.979 So, we'll we'll try to catch that carefully. Nate, did you want to say something? I was just gonna say I think if we remove non substantive or without something else to to qualify it, it becomes so broad that it competes with the prior sentence, that paragraph requires.

1106 "" (0) 06:49:27.979 --> 06:49:43.769 Well, I think the the accessibility point is the narrow, is the well the most importantly narrowing piece of this. Is it though because it I mean it's that can be very broad, like in excerps. Excepts also could be narrowing, but excepts kind of thing could be everything.

1107 "" (0) 06:49:43.769 --> 06:50:03.029 So yeah, I I think, yeah, I, if accessibility just means like accessibility could be you've summarized my entire thing into a summary because I mean isn't it the whole point of like the sort of AI service summaries is that it makes the content more accessible on the page, right? Like.

1108 "" (0) 06:50:03.029 --> 06:50:21.869 I I feel like this paragraph without that qualifier directly competes from previous like they contradict each other. So some if it's not sound substantive, it'd be something else, I also think alterations could work in with respect to changes, but I maybe that's some drafting stuff there, but I do think.

1109 "" (0) 06:50:21.869 --> 06:50:38.279 That would be opportunity. Let's do this. Let's let's break for the day. People can go and and have a bit more of a think and asleep about it. We'll, we'll briefly bring it back up again tomorrow morning to see if people have had further thoughts about it. And if if.

1110 "" (0) 06:50:38.279 --> 06:50:57.599 That goes well, we'll we'll take it to the list, I think. And tomorrow I think the things that we have on the agenda are going back to three.two and also back to the training term, not necessarily in that order. Yep. Okay, ok.

1111 "" (0) 06:50:59.099 --> 06:51:19.099 All right, let's call today. Thank you. Thanks, everybody. Thank you.