File:August 2024 Future Audiences office hours.webm

00:00:00.000 --> 00:00:00.650 Liam Wyatt (WMF): ok.

2 00:00:02.780 --> 00:00:09.957 Maryana Pinchuk: Alright. Thank you. Hello, everyone. I am Mariana Pinchuk. I am the lead of future audiences.

3 00:00:10.440 --> 00:00:16.489 Maryana Pinchuk: and we are going to be giving our monthly updates to you all.

4 00:00:16.869 --> 00:00:30.860 Maryana Pinchuk: As you're joining in the chat. Thank you for those who are already doing so. Please introduce yourselves. Drop a note about where you're where you're coming from, what projects you work on. If this is your 1st time and then a future audiences call, please let us know.

5 00:00:31.010 --> 00:00:34.609 Maryana Pinchuk: and I will bring up our agenda

6 00:00:35.010 --> 00:00:36.769 Maryana Pinchuk: here in a second.

7 00:00:37.400 --> 00:00:38.060 Maryana Pinchuk: See?

8 00:00:38.860 --> 00:00:39.810 Maryana Pinchuk: Go.

9 00:00:40.480 --> 00:00:41.400 Maryana Pinchuk: Okay.

10 00:00:42.930 --> 00:01:11.139 Maryana Pinchuk: so yes, please drop your note in the chat, and we'll have some time for QA. Both around the new experiment that we're working on, which we'll get to in a minute and hopefully some time for open discussion on any topics you want to bring up. So please save your questions, for then. But if you do have thoughts that you want to share, you want to forget we have a notes document that you're welcome to drop your notes onto.

11 00:01:12.160 --> 00:01:15.800 Maryana Pinchuk: Mr. SJ. I'm gonna mute. You cause I can hear your keyboard.

12 00:01:16.710 --> 00:01:25.630 Maryana Pinchuk: There we go. Perfect. All right. Welcome, everyone. Thanks for joining us. I'm Mariana Pinchuk. I am the lead on future audiences.

13 00:01:25.630 --> 00:01:51.689 Maryana Pinchuk: This is our agenda. Today. We are going to every call. We try to recontextualize what future audiences is in case you're new here, going to give you an update on the last fiscal year, which was the 1st year that this team existed very exciting, and give you some insights on our last experiment that we performed citation needed, which we demoed in this very venue in, I believe, April.

14 00:01:51.910 --> 00:02:03.680 Maryana Pinchuk: And then we're gonna talk about a new experiment that we're working on. So I'll turn it over to my colleague Francisco, who's here? To give you a walkthrough and we'll have some time for questions and feedback

15 00:02:04.023 --> 00:02:19.859 Maryana Pinchuk: and also we're gonna be asking you all to test this thing out for us, please. So we'll have some instructions for how you can get access to it. And then, like I said at the end. Any other questions you have? Please. Feel free. We have time. We'll try to get to them.

16 00:02:20.220 --> 00:02:48.600 Maryana Pinchuk: So just to start future audiences. What is future audiences? It is an initiative of the product and technology department at the Wikimedia Foundation. We've been around for about a year, and we have a mandate to test new ways of serving future generations of knowledge, consumers and contributors in new ways and learning. What can be done differently? What kinds of

17 00:02:48.940 --> 00:02:56.209 Maryana Pinchuk: features or technologies, or just ways of approaching knowledge, sharing and knowledge. Consumption

18 00:02:56.210 --> 00:03:19.779 Maryana Pinchuk: can be brought into our movement and where we can go next. But our team doesn't build products like a lot of the other product and tech teams, which is a little confusing. I know we build experiments. We build experimental features or tools that are live for a temporary period of time in order to learn something valuable in

19 00:03:19.780 --> 00:03:36.679 Maryana Pinchuk: whatever space we're testing in. And the ultimate goal of this is to make recommendations for bigger investments that all the other product and technology teams should be making to to help us really continue to evolve as a movement to continue to serve the needs of

20 00:03:36.680 --> 00:03:49.099 Maryana Pinchuk: everyone in the world. With free knowledge. So like, I said, we've been around for about a year and our fiscal year at the Wikimedia Foundation. Starts in July. So we're in

21 00:03:49.100 --> 00:04:12.839 Maryana Pinchuk: the very beginning of the new fiscal year, and we did a lot of experiments last year which we've shared in this venue. And if you want to go and read about all of them on Meta, go to future audiences. All of them are listed there, but our goal is to make strategic recommendations in the kind of draft annual planning period which runs from April through June.

22 00:04:13.335 --> 00:04:30.409 Maryana Pinchuk: So I'll just give you a quick rundown of everything we learned last year. The recommendations or insights that we brought into the annual planning period this past April, June. And what we're we're working on next in this new fiscal year which just kicked off

23 00:04:31.082 --> 00:04:39.279 Maryana Pinchuk: so the 1st thing we learned is that we can, in fact, run quick experiments and learn and turn things off

24 00:04:39.809 --> 00:05:03.190 Maryana Pinchuk: which is not necessarily the the way that we approach sort of standard product building at the Wikipedia Foundation. All of the other product teams 1st of all, are much larger. They have more more resources for software engineers and designers and user research. And it's a big cross functional team usually working on a a very

25 00:05:03.190 --> 00:05:14.329 Maryana Pinchuk: complex involved project that has to think about things like multi-language support. So we can't just build things that work for English Wikipedia. We have to build things that work for every language. Wikipedia.

26 00:05:14.330 --> 00:05:41.659 Maryana Pinchuk: We have to consider things like the kinds of browsers that people are using accessibility, usability. All kinds of considerations have to go into making real products that are fully scaled to meet the needs of our current audiences. But our team is different. We're not trying to build new products. We're trying to learn quickly and learn about things that are very emergent and new and changing very rapidly, such as AI.

27 00:05:41.660 --> 00:05:55.490 Maryana Pinchuk: So last year we ran a lot of small scale. AI experiments starting with a Plugin for Chatgpt that brought information from Wikipedia to Chatgpt users in a kind of a new way.

28 00:05:55.750 --> 00:06:18.090 Maryana Pinchuk: and we ran that experiment learned some stuff and we turned it off. We did not invest in continuing to support that or build that out further, because that is not the role of our team and a big learning, in addition to just learning that we could, in fact, run quick experiments, learn and get what we needed and not have to preserve things forever if they weren't needing

29 00:06:18.448 --> 00:06:34.950 Maryana Pinchuk: the needs of users in real life. What? We really explored a lot around and and learned a lot about was AI the capabilities of AI. And you know, when Chatgbt 1st came around, everyone was really excited, and there were a lot of.

30 00:06:34.980 --> 00:06:50.919 Maryana Pinchuk: you know, probably a little over hyped press pieces on this idea that oh, AI is just gonna generate all of this content. It's just maybe it'll just write Wikipedia for us, and we won't even need humans anymore, because it'll be so good. And it'll just create all this new content.

31 00:06:51.480 --> 00:07:08.369 Maryana Pinchuk: we have not so far found that that is a reality with the tools that exist today. However, one thing that we kept learning, we learned this through the Chat Gpt Plugin, and also through citation needed, which I'll talk about in a minute. Another experiment that we ran pretty quickly.

32 00:07:08.764 --> 00:07:20.420 Maryana Pinchuk: Was that AI? The the thing about AI that's really interesting is that it can really help to kind of augment the navigation and parsing of a lot of really complex content.

33 00:07:20.735 --> 00:07:44.370 Maryana Pinchuk: Which, if you've ever tried to read a long Wikipedia article or tried to search Wikipedia for something you will know that that is a a big problem that exists on our projects today. We've generated so much content over the last 23 years as a movement. And it's big and long and all over the place, and not structured necessarily in the way that people can navigate it easily.

34 00:07:44.460 --> 00:08:05.509 Maryana Pinchuk: So we see a lot of opportunities for using AI in this particular way, not to create new stuff necessarily, but to really figure out how to get people to the information that they need and help them make sense of it. Better so for the other product team. Some of the stuff that they're going to be working on this year. Starting now in July.

35 00:08:05.903 --> 00:08:18.489 Maryana Pinchuk: Are things like researching and prototyping new ways that AI could be used to facilitate better content discovery and browsing on our projects. And for the contributor side.

36 00:08:18.490 --> 00:08:25.240 Maryana Pinchuk: trying to see if there are ways that we can train Llms to detect known article issues. For example.

37 00:08:25.240 --> 00:08:53.999 Maryana Pinchuk: Peacock language. This new product that was released is the best product in the world. This amazing revolutionary new product that's going to change everything was released on January 20. You know the drill about peacock language. So we're trying to see if we can use AI tools to detect that kind of article issue and then serve that through existing experiences geared towards our contributors, such as edit check and suggested edits

38 00:08:54.517 --> 00:09:11.480 Maryana Pinchuk: and the out of fact experiment which I will be talking about in a second is another way in which we think we want to learn more about how AI can be used to facilitate parsing through a bunch of content on Wiki off Wiki, making sense of it and helping an editor to make a decision.

39 00:09:12.890 --> 00:09:30.790 Maryana Pinchuk: So citation needed. This was the last experiment that we conducted, and we ran this last couple of quarters. We just wrapped it up now, and there's a report on Wiki that if you want to read more about all of the things that we found through the course of this experiment. It's really interesting. Project.

40 00:09:31.060 --> 00:09:35.229 Maryana Pinchuk: Just as a reminder. If you weren't here in in April, when we talked about this

41 00:09:35.580 --> 00:09:51.510 Maryana Pinchuk: our goal was to see if we could use Wikipedia, plus the kind of searching, retrieving power of AI to allow people to verify content on websites that they were getting information from online.

42 00:09:51.510 --> 00:10:09.240 Maryana Pinchuk: And we didn't know if people would be interested in this at all. Certainly Wikipedians are interested in this, but when it came to kind of the broader public we weren't sure if this is a behavior that people wanted to do if they would trust Wikipedia as a resource in verifying content.

43 00:10:09.575 --> 00:10:15.560 Maryana Pinchuk: If we could reach new audiences, and if if AI could actually facilitate this at all. These were all

44 00:10:15.630 --> 00:10:42.160 Maryana Pinchuk: big, open questions. As we embarked on building a browsing plugin that would allow you to essentially, highlight claims on websites. Such as the one that you're seeing on the screen on the right here. So you can highlight this claim on any website that you find. And you. Citation needed to check to see what Wikipedia has to say about this claim, whether it's already there, and and there's a reference, or whether it's

45 00:10:42.210 --> 00:10:46.530 Maryana Pinchuk: not on Wikipedia, or whether Wikipedia has something that contradicts that claim.

46 00:10:47.697 --> 00:10:59.500 Maryana Pinchuk: So what we found when we built this, Plugin launched it, put it in the chrome store publicized it in various kind of small lightweight ways. We

47 00:10:59.820 --> 00:11:05.610 Maryana Pinchuk: we wrote a blog post that mentioned it. We asked our sort of staff members with a lot of

48 00:11:05.620 --> 00:11:20.940 Maryana Pinchuk: clout on social media to publicize it among their friends and peers. We emailed some donors who had been interested in kind of keeping tabs with what the Wikimedia Foundation is doing. To just let them know about this experiment.

49 00:11:21.251 --> 00:11:40.878 Maryana Pinchuk: And what we got back was an interesting mix of insights. So, on the one hand, people love the idea. We got lots and lots of great feedback from users from all over the place. That this is a really cool idea really needed. Really, like the concept a lot. And and people did over a thousand people took the

50 00:11:41.300 --> 00:12:02.260 Maryana Pinchuk: took the extra step of actually installing it and trying it out. So they didn't just say that they liked the idea. They they installed it in in their chrome web browser, which is a pretty big step, I mean browser. Browser extensions are kind of finicky, not not super easy to to use. So that was great to know that there, there's something here.

51 00:12:02.260 --> 00:12:13.039 Maryana Pinchuk: But what we saw is that you know we would let people know about this thing. They would test it out, and they didn't keep using it. After that they would use it once or twice, and then they kind of went away.

52 00:12:13.290 --> 00:12:38.679 Maryana Pinchuk: So I think there's a lot more we can learn about how to actually build this in a way that delivers value to these people who really want something like this to exist, and there are probably a lot of different approaches we could take, whether it's the kind of content that this is being presented on how it's being presented. How manual it is versus something that kind of just runs in the background and checks for incorrect claims.

53 00:12:39.027 --> 00:13:04.029 Maryana Pinchuk: But 1 1 kind of next step we're thinking about is showing this to 3rd party platforms who we know might have some issues with not super reliable content appearing on on their their apps, their their services. And really seeing if if we can continue to learn through putting this in a place where it could be a little bit more intuitive and easy to use

54 00:13:04.459 --> 00:13:12.259 Maryana Pinchuk: and work a little bit more specifically rather than generally. So that's 1 thing we're gonna we're gonna be working on

55 00:13:13.158 --> 00:13:22.750 Maryana Pinchuk: another thing that we learned, though, super super crucial. And this really gets back into the next experiment that we'll talk about, and also our overall

56 00:13:22.890 --> 00:13:24.929 Maryana Pinchuk: findings about AI

57 00:13:25.790 --> 00:13:26.639 Maryana Pinchuk: Llms

58 00:13:26.780 --> 00:13:42.100 Maryana Pinchuk: can be really helpful in finding and analyzing and retrieving information, especially if you have a huge messy corpus of information as the Committee of Projects do, but they're not perfect by any means. They make mistakes. They make stuff up.

59 00:13:42.120 --> 00:13:47.099 Maryana Pinchuk: They don't have the kind of context and judgment that human does

60 00:13:47.431 --> 00:14:09.640 Maryana Pinchuk: and for sure there could have been a lot more prompt engineering and sort of careful refinement of the model that we were using, and citation needed to try to minimize those mistakes. But we feel that at the end of the day it's it's never going to be perfect. And there are just some things that human judgment is really really required for. And this this is it.

61 00:14:09.640 --> 00:14:27.020 Maryana Pinchuk: this being really analyzing facts that can be complex and nuanced and multifaceted. We're not in a world in which AI can replace a human being to really make a good judgment about the factuality of something, or how good the reference is.

62 00:14:27.380 --> 00:14:47.940 Maryana Pinchuk: So so that was a really interesting finding. And in that vein, we're gonna talk a little bit about a new experiment that we're starting and I'm gonna hand it over to my colleague Francisco, to talk about this and Llm. Human in the loop off platform contribution experiments. Take it away, Francisco.

63 00:14:47.940 --> 00:14:53.090 Francisco Navas: Alright! Alright! I assume you can hear me. Maybe someone say Yes.

64 00:14:53.770 --> 00:14:54.620 Liam Wyatt (WMF): Yes.

65 00:14:54.750 --> 00:14:59.174 Francisco Navas: Thank you very much. Look at those thumbs. Yeah, hey, folks, I'm from Cisco.

66 00:15:00.600 --> 00:15:18.010 Francisco Navas: Just been on a future island for a little bit, now jumped on to work on this project super exciting and out of fact, as you can see it. As Mariana was saying, it very much comes from some of these learnings about citation needed. Liam wrote. Knowledge is human, Mariana said.

67 00:15:18.420 --> 00:15:23.519 Francisco Navas: Is the Lm really is AI at all, or Lm. As a subset

68 00:15:23.700 --> 00:15:32.039 Francisco Navas: strong enough to tell what is factual or not. Well, I I think if we say no, then we can use the human to do something else.

69 00:15:33.310 --> 00:15:44.279 Francisco Navas: it's always the work of Wikipedians to find information and collate it, collect it, rewrite it, organize it, judge it, and put it into Wiki where is necessary.

70 00:15:44.300 --> 00:15:50.529 Francisco Navas: So we are just at the point of having an Mvp. Ready for. Out of fact.

71 00:15:51.370 --> 00:15:54.289 Francisco Navas: I tried to put the basics down here in this slide

72 00:15:54.310 --> 00:16:04.319 Francisco Navas: you will recognize it looks a lot like citation needed. Of course, it's built off of the framework and the same Llm. Engine, and even

73 00:16:04.640 --> 00:16:07.040 Francisco Navas: tweaked, but very similar.

74 00:16:08.030 --> 00:16:19.450 Francisco Navas: of course. Maybe Daniel, our main engineer on this project, may differ, and he will have more specific but very similar engine for telling the user who selected a statement

75 00:16:19.650 --> 00:16:27.339 Francisco Navas: what is going on in the Wikipedia article. So the main difference, of course, right between citation needed and that of fact is, maybe citation needed

76 00:16:27.420 --> 00:16:30.330 Francisco Navas: was trying to help. You understand if

77 00:16:30.440 --> 00:16:36.030 Francisco Navas: what was on what the text you said selected, if it was on the Wikipedia article.

78 00:16:36.040 --> 00:16:41.849 Francisco Navas: And so we're using that function now without a fact to help you decide whether a statement you selected.

79 00:16:41.860 --> 00:16:54.299 Francisco Navas: if whether it's on or not on or how much the article agrees with that statement you selected. If that selected text belongs on an article that you that you chose.

80 00:16:54.870 --> 00:16:57.019 Francisco Navas: And so, as you can see here

81 00:16:57.310 --> 00:16:59.400 Francisco Navas: we use Lm to help you

82 00:16:59.480 --> 00:17:02.459 Francisco Navas: speed up that process or help you make that some decisions.

83 00:17:03.905 --> 00:17:04.910 Francisco Navas: And

84 00:17:05.190 --> 00:17:11.700 Francisco Navas: you can see how it works here, and I think the next slide means that we have to try some live demo. Okay, we'll get to live, Demo in a second

85 00:17:12.316 --> 00:17:14.700 Francisco Navas: like, with all these experiments.

86 00:17:15.430 --> 00:17:21.800 Francisco Navas: They're specific question based. We have some hypotheses we want to check on

87 00:17:23.040 --> 00:17:25.250 Francisco Navas: Does. Has there ever been?

88 00:17:25.500 --> 00:17:26.710 Francisco Navas: I don't think so.

89 00:17:27.243 --> 00:17:33.159 Francisco Navas: A productive way to contribute from, not Wikipedia org onto Wikipedia org.

90 00:17:33.580 --> 00:17:37.210 Francisco Navas: Can we try that? What results does that produce?

91 00:17:37.822 --> 00:17:39.570 Francisco Navas: That it obviously

92 00:17:39.740 --> 00:17:45.059 Francisco Navas: raises the question of, who could that be for? That's the 3rd question here is that.

93 00:17:45.420 --> 00:17:58.190 Francisco Navas: could there be a tool for productively contributing to Wikipedia? Does that maybe go to, or should be used by non Wikipedia editors, people who don't have experience, your norm, as I call them.

94 00:17:59.550 --> 00:18:08.080 Francisco Navas: simultaneously. Of course, like we have to understand if having the AI 11 in the loop is actually helpful, does it harm the process?

95 00:18:10.060 --> 00:18:12.733 Francisco Navas: do we mean? What can we do here? Yep,

96 00:18:13.560 --> 00:18:14.889 Francisco Navas: it's a good question.

97 00:18:15.470 --> 00:18:22.339 Francisco Navas: I contribute often Wikipedia, through editing Wikiped with data. I only say Wikipedia here, because

98 00:18:22.500 --> 00:18:24.339 Francisco Navas: so far out of fact.

99 00:18:24.540 --> 00:18:29.820 Francisco Navas: is just for Wikipedia in particular English Wikipedia, at least for this Mvp launch.

100 00:18:29.840 --> 00:18:31.010 Francisco Navas: And

101 00:18:31.210 --> 00:18:34.190 Francisco Navas: but yeah, why not the Wikimedia

102 00:18:34.390 --> 00:18:38.500 Francisco Navas: editor world? So the non Wikimedian, I will say

103 00:18:39.730 --> 00:18:41.789 Francisco Navas: And then, finally, a super important question is.

104 00:18:41.920 --> 00:18:47.849 Francisco Navas: what about sources? How can a tool produced by Wmf

105 00:18:49.660 --> 00:18:50.960 Francisco Navas: support

106 00:18:51.270 --> 00:18:55.279 Francisco Navas: source decisions across different Wiki projects?

107 00:18:55.290 --> 00:18:57.099 Francisco Navas: And what place. Does

108 00:18:57.440 --> 00:19:00.049 Francisco Navas: a tool have been doing that?

109 00:19:00.460 --> 00:19:05.659 Francisco Navas: You know, it's all these questions about what it should be automated and what should not be automated.

110 00:19:06.351 --> 00:19:10.659 Francisco Navas: Cool details. Important details like I said, only on English wiki.

111 00:19:10.730 --> 00:19:13.289 Francisco Navas: and it has a post limit per day

112 00:19:14.590 --> 00:19:17.599 Francisco Navas: trying to stop some spam. You could easily see how

113 00:19:17.830 --> 00:19:21.500 Francisco Navas: someone could just spam the heck out of facts. We don't want that.

114 00:19:22.090 --> 00:19:27.800 Francisco Navas: The 1st step we took for reliability. Daniel implemented the head bomb gadget

115 00:19:27.930 --> 00:19:29.060 Francisco Navas: to

116 00:19:29.070 --> 00:19:33.759 Francisco Navas: create a little signal. Once a text from a source is selected about

117 00:19:33.840 --> 00:19:37.409 Francisco Navas: where that source might sit on the perennial sources list.

118 00:19:37.720 --> 00:19:44.429 Francisco Navas: And finally, but very important. Thank you. To Alana and Sj. Who are here, who

119 00:19:44.470 --> 00:19:49.890 Francisco Navas: birth this idea? We've sat for 12 h on a train going to Toronto.

120 00:19:50.270 --> 00:19:58.780 Francisco Navas: The 12 h was a mistake, but hanging out with them was surely not. We would not have even started this idea had it not been for them. So

121 00:19:58.860 --> 00:20:07.750 Francisco Navas: thank you very much for your help, support, and general mentorship. And so here we are. So those are some important basic facts.

122 00:20:08.590 --> 00:20:10.539 Francisco Navas: And I think we could do a live demo now.

123 00:20:11.360 --> 00:20:12.450 Francisco Navas: Daniel.

124 00:20:12.840 --> 00:20:13.810 Francisco Navas: can we

125 00:20:14.090 --> 00:20:15.049 Francisco Navas: think we can?

126 00:20:15.300 --> 00:20:19.069 Francisco Navas: I think we're confident we're confident. I'm happy to run it.

127 00:20:21.390 --> 00:20:22.020 Daniel Erenrich: Sure.

128 00:20:22.770 --> 00:20:23.550 Francisco Navas: Okay.

129 00:20:23.680 --> 00:20:25.900 Francisco Navas: let's see what it

130 00:20:26.330 --> 00:20:29.270 Francisco Navas: does when I try to share my screen.

131 00:20:30.300 --> 00:20:32.800 Francisco Navas: Oh, some privacy problem.

132 00:20:32.860 --> 00:20:34.110 Francisco Navas: Hmm.

133 00:20:34.630 --> 00:20:36.810 Francisco Navas: zoom, yes.

134 00:20:37.515 --> 00:20:42.399 Maryana Pinchuk: Liam, maybe you need to make Francisco co-host so he can share his screen. Possibly.

135 00:20:42.480 --> 00:20:44.270 Liam Wyatt (WMF): He he already is.

136 00:20:44.270 --> 00:20:45.460 Maryana Pinchuk: Oh hmm!

137 00:20:46.880 --> 00:20:48.220 Francisco Navas: About now.

138 00:20:49.462 --> 00:20:51.000 Francisco Navas: But papa, papa!

139 00:20:51.350 --> 00:20:53.200 Liam Wyatt (WMF): That's what you get for the live Demo.

140 00:20:53.410 --> 00:20:57.829 Francisco Navas: There we go, Daniel. I think you gotta do it, cause I can't share

141 00:20:57.900 --> 00:20:59.360 Francisco Navas: my screen on Zoom

142 00:20:59.660 --> 00:21:00.460 Francisco Navas: apologies.

143 00:21:00.460 --> 00:21:07.039 Liam Wyatt (WMF): Think this is not. This is not a zoom privacy problem. This is your computer needs configuration to zoom. Yeah.

144 00:21:08.498 --> 00:21:13.863 Daniel Erenrich: Okay, I hadn't been anticipating demoing. So this is gonna be interesting. Oh, wait! Do I have the same problem

145 00:21:14.690 --> 00:21:15.889 Francisco Navas: Oh, I got it!

146 00:21:16.190 --> 00:21:16.690 Daniel Erenrich: Okay.

147 00:21:16.690 --> 00:21:17.670 Maryana Pinchuk: Screen, alright.

148 00:21:17.670 --> 00:21:19.970 Liam Wyatt (WMF): Worrying, confirming. I can see your screen.

149 00:21:19.970 --> 00:21:20.620 Francisco Navas: Great.

150 00:21:20.840 --> 00:21:22.880 Francisco Navas: All right, we're on wikipedia.com

151 00:21:23.000 --> 00:21:25.670 Francisco Navas: wikipedia.com. I always say that wikipedia.org

152 00:21:27.340 --> 00:21:31.219 Francisco Navas: And here's the extension with a very cool dark.

153 00:21:31.500 --> 00:21:33.239 Francisco Navas: This is the dev mode.

154 00:21:33.810 --> 00:21:35.070 Francisco Navas: So right now

155 00:21:35.800 --> 00:21:37.558 Francisco Navas: try and run it on.

156 00:21:37.960 --> 00:21:41.910 Francisco Navas: Any Wikipedia page, Wikipedia Page, or the extension

157 00:21:42.420 --> 00:21:45.440 Francisco Navas: a chrome extension itself. It won't work.

158 00:21:45.610 --> 00:21:47.059 Francisco Navas: So first, st things 1st

159 00:21:47.220 --> 00:21:48.620 Francisco Navas: prompts me to log in.

160 00:21:51.260 --> 00:21:54.560 Francisco Navas: So I will do that hopefully. I can log in no problem.

161 00:21:54.970 --> 00:21:56.500 Francisco Navas: Don't look at my password.

162 00:22:02.100 --> 00:22:03.650 Francisco Navas: Okay, that's the test.

163 00:22:05.140 --> 00:22:07.740 Francisco Navas: Alright. So I should be logged in now.

164 00:22:08.590 --> 00:22:10.259 Francisco Navas: So now oop

165 00:22:10.470 --> 00:22:13.080 Francisco Navas: trying to move this guy down here.

166 00:22:13.896 --> 00:22:24.159 Francisco Navas: Okay, so say you're a reader of infowars, and for some reason you think that infowars belongs. Information from infowars belongs on Wikimedia

167 00:22:24.170 --> 00:22:25.830 Francisco Navas: on any Wiki project.

168 00:22:26.270 --> 00:22:28.819 Francisco Navas: Please hold, folks, I'm being facetious.

169 00:22:30.690 --> 00:22:32.020 Francisco Navas: and come down here.

170 00:22:32.270 --> 00:22:37.789 Francisco Navas: She's a week just through the headline. So, according to worse, when us universities to require covid vaccine.

171 00:22:40.060 --> 00:22:40.970 Francisco Navas: Okay.

172 00:22:41.420 --> 00:22:44.900 Francisco Navas: so artifact immediately tells you the source is unreliable.

173 00:22:44.950 --> 00:22:46.879 Francisco Navas: If we were to hit, learn more.

174 00:22:47.390 --> 00:22:50.859 Francisco Navas: take us through the reliability learning Sources list.

175 00:22:51.410 --> 00:22:54.390 Francisco Navas: The idea is, of course, if you're not someone

176 00:22:54.480 --> 00:22:56.110 Francisco Navas: who edits Wiki

177 00:22:56.720 --> 00:22:58.290 Francisco Navas: can help you learn about

178 00:22:58.390 --> 00:23:03.039 Francisco Navas: what volunteers have decided around sources

179 00:23:03.640 --> 00:23:06.220 Francisco Navas: for stretches, we'll check the statement.

180 00:23:07.000 --> 00:23:17.279 Francisco Navas: So it's not gonna tell you, if the statement is true or false, it's not the point here what what it can the Lm. Is doing right here is analyzing the presence of the statement on

181 00:23:17.490 --> 00:23:21.040 Francisco Navas: different English Wikipedia articles

182 00:23:21.750 --> 00:23:26.780 Francisco Navas: and trying to help you understand what it mentions. And then it makes an assessment.

183 00:23:27.100 --> 00:23:33.879 Francisco Navas: So, for example, the source, mentions that many us colleges and universities does not have required, but does not provide the exact number. So okay.

184 00:23:34.100 --> 00:23:35.110 Francisco Navas: that's fine.

185 00:23:35.560 --> 00:23:42.170 Francisco Navas: That's a good. That's a pretty good one. Here's just some other relevant ones just about us colleges and COVID-19 in general.

186 00:23:42.290 --> 00:23:44.369 Francisco Navas: Again, this one is less specific.

187 00:23:44.430 --> 00:23:49.720 Francisco Navas: So maybe these are. You think these are not relevant at all. And actually you may want to add this

188 00:23:49.890 --> 00:23:51.520 Francisco Navas: back to the infowars

189 00:23:52.300 --> 00:23:53.250 Francisco Navas: article

190 00:23:53.660 --> 00:23:56.030 Francisco Navas: so you can search on Wiki.

191 00:23:56.580 --> 00:23:58.100 Francisco Navas: Your search will. Then

192 00:23:59.180 --> 00:24:02.760 Francisco Navas: the Lm. Will then again search through that article that you chose.

193 00:24:02.820 --> 00:24:06.090 Francisco Navas: And yeah, let's let's put this on there, for now let's see.

194 00:24:06.440 --> 00:24:15.489 Francisco Navas: I should say that anything coming from the dev version about a fact will add to we'll add the fact to

195 00:24:16.140 --> 00:24:21.210 Francisco Navas: the test Wiki version of the article, so I am not putting on this on Wikipedia right now.

196 00:24:21.380 --> 00:24:22.419 Francisco Navas: Don't wear it

197 00:24:23.536 --> 00:24:26.389 Francisco Navas: so we generate this form. You can link

198 00:24:26.550 --> 00:24:28.100 Francisco Navas: to the article itself.

199 00:24:28.300 --> 00:24:31.479 Francisco Navas: So you want to read it for any more context that you may need.

200 00:24:31.980 --> 00:24:36.089 Francisco Navas: And and this is a form that shows you what the post

201 00:24:36.400 --> 00:24:38.009 Francisco Navas: to a

202 00:24:38.520 --> 00:24:43.850 Francisco Navas: a top page for this article in this case. To the test Wiki would look like.

203 00:24:43.860 --> 00:24:45.589 Francisco Navas: and you can add your additional comment.

204 00:24:47.460 --> 00:24:48.800 Francisco Navas: truly think

205 00:24:48.940 --> 00:24:50.170 Francisco Navas: it's important

206 00:24:50.710 --> 00:24:55.460 Francisco Navas: to add info about Covid to the worst

207 00:24:55.640 --> 00:24:56.380 Francisco Navas: article.

208 00:24:57.540 --> 00:24:58.779 Francisco Navas: Okay, great.

209 00:25:01.680 --> 00:25:04.679 Francisco Navas: So now it's been sent to the top page for infowars.

210 00:25:05.580 --> 00:25:07.150 Francisco Navas: You can see in your new tab.

211 00:25:07.990 --> 00:25:10.879 Francisco Navas: and it automatically automatically generates this

212 00:25:11.560 --> 00:25:12.589 Francisco Navas: this post.

213 00:25:12.900 --> 00:25:17.679 Francisco Navas: Here's an interesting article, because normally

214 00:25:18.241 --> 00:25:26.390 Francisco Navas: out of fact, would use Sito to create a reference that you can then post, but I guess it doesn't work on infowars. That's new information to me.

215 00:25:26.510 --> 00:25:29.799 Francisco Navas: So why don't we try it on a different site?

216 00:25:30.240 --> 00:25:32.039 Francisco Navas: Zion Kamal Harris

217 00:25:32.600 --> 00:25:38.420 Francisco Navas: article. It was an interesting point here about the importance of Michigan and statistics.

218 00:25:39.870 --> 00:25:41.400 Francisco Navas: just to run it again.

219 00:25:43.430 --> 00:25:47.730 Francisco Navas: I think it's very interesting and very valuable to read the

220 00:25:47.900 --> 00:25:50.870 Francisco Navas: assessments that the Llm. Makes

221 00:25:50.900 --> 00:25:53.299 Francisco Navas: and could be helpful in choosing an article.

222 00:25:53.840 --> 00:26:03.270 Francisco Navas: For example, here the claim is partially correct. Hmm. As Democrats did take control of both Houses and the Machine Legislature for the 1st time in 45 years. But it's incorrect about 2018

223 00:26:03.420 --> 00:26:12.980 Francisco Navas: midterms and 2020 Presidential elections. So again. Mvp. I'm don't love correct and incorrect here, but what I think it's trying to say is that in this article

224 00:26:13.170 --> 00:26:15.240 Francisco Navas: those specific points are not present.

225 00:26:15.960 --> 00:26:18.520 Francisco Navas: So that will be something to correct, to correct.

226 00:26:19.920 --> 00:26:21.360 Francisco Navas: Here's a sit toy test.

227 00:26:22.390 --> 00:26:25.249 Francisco Navas: So site. So it should generate a

228 00:26:28.930 --> 00:26:29.810 Francisco Navas: here we go.

229 00:26:30.640 --> 00:26:32.149 Francisco Navas: And so the idea is.

230 00:26:33.410 --> 00:26:36.549 Francisco Navas: there's a signed by me, added by artifact.

231 00:26:37.600 --> 00:26:41.899 Francisco Navas: creates a subject for the new topic.

232 00:26:42.410 --> 00:26:44.440 Francisco Navas: And ideally.

233 00:26:44.590 --> 00:26:49.149 Francisco Navas: if this was a fact that folks on this page wanted to add.

234 00:26:49.370 --> 00:26:54.999 Francisco Navas: they could argue it here, discuss it. Take it from, take this information and make it into an actual edit

235 00:26:55.687 --> 00:26:57.910 Francisco Navas: and then ideally use the reference as A

236 00:26:58.060 --> 00:26:59.799 Francisco Navas: into the article itself.

237 00:26:59.920 --> 00:27:03.519 Francisco Navas: So the idea, of course, as I said, is to support

238 00:27:04.000 --> 00:27:05.220 Francisco Navas: adding facts

239 00:27:05.610 --> 00:27:07.990 Francisco Navas: into part of the

240 00:27:09.220 --> 00:27:10.300 Francisco Navas: from off

241 00:27:10.820 --> 00:27:12.190 Francisco Navas: wikipedia.org

242 00:27:13.011 --> 00:27:14.980 Francisco Navas: and seeing what we learn from there.

243 00:27:15.481 --> 00:27:18.199 Francisco Navas: I think there was maybe one or 2 more slides.

244 00:27:18.250 --> 00:27:21.400 Francisco Navas: and then we'll open up some questions. Should I look, be looking at the chat.

245 00:27:22.086 --> 00:27:27.293 Maryana Pinchuk: I'm I'm kind of trying to drop some notes in here under the chat as well.

246 00:27:27.690 --> 00:27:29.697 Maryana Pinchuk: yeah, Mike, you asked.

247 00:27:31.088 --> 00:27:55.240 Maryana Pinchuk: given that we know that some talk pages can be pretty inactive. Would it help to show kind of when the talk page was last edited? Yes, absolutely. I think that the talk page posting is kind of our very simple proof of concept, the implementation of something like this. But we know that there are. You know, there are lots of unfortunately, pages that aren't monitored very closely and talk pages that aren't necessarily acted on so

248 00:27:55.240 --> 00:28:14.479 Maryana Pinchuk: looking for really interested in hearing ideas about other places that the fact could go. As you can see, we're very hesitant to add anything directly to Wikipedia, because we're trying to keep a kind of like a open funnel, but but but only very you know, careful actual posting to project model.

249 00:28:14.480 --> 00:28:34.079 Maryana Pinchuk: But if you have other ideas like, I don't know. Wiki projects maybe associated with the article, or maybe an entirely new queue of facts that can be reviewed by experienced editors really love to hear thoughts on that. I'm gonna share my screen again, if I can find it. Where is it?

250 00:28:34.600 --> 00:28:35.670 Maryana Pinchuk: There go.

251 00:28:36.800 --> 00:28:37.430 Maryana Pinchuk: Okay.

252 00:28:37.430 --> 00:28:39.129 Francisco Navas: Carolina for the props

253 00:28:42.140 --> 00:28:51.679 Francisco Navas: so hopefully that was smooth and not horrible. To listen to me talk through this for a while. I couldn't see your faces because I was looking at the screen so I couldn't judge. But anyway.

254 00:28:51.850 --> 00:28:55.096 Francisco Navas: you are here. Thank you for coming

255 00:28:56.240 --> 00:28:57.810 Francisco Navas: We'd love for your feedback.

256 00:28:57.830 --> 00:28:59.350 Francisco Navas: There's a couple ways

257 00:28:59.680 --> 00:29:01.970 Francisco Navas: number one, just from this demo.

258 00:29:02.080 --> 00:29:06.570 Francisco Navas: You can, you know. Text me if you have my number, email, me post on the middle talk page.

259 00:29:06.890 --> 00:29:14.630 Francisco Navas: There's the link. It's pretty easy to find you try. This is that a fact comes up on Google as well? We can admit

260 00:29:14.680 --> 00:29:24.470 Francisco Navas: we can send it to you all later as well. If we have permission to email you and then, of course, we would love if you could trial our dev version.

261 00:29:24.570 --> 00:29:26.510 Francisco Navas: we have a

262 00:29:27.230 --> 00:29:32.500 Francisco Navas: it's running, and we have a goal of getting a better version out ideally for Wikimania.

263 00:29:33.200 --> 00:29:37.060 Francisco Navas: In Poland, where Mariana will be in a daily presenting the

264 00:29:37.110 --> 00:29:49.650 Francisco Navas: prod prod version that totally works. And it's incredible. And so we yeah, we'd love all your feedback for that use. You'll need Nikki an end, Wiki, and test Wiki account and auto confirmation status.

265 00:29:50.430 --> 00:29:54.850 Francisco Navas: I did not have test Wiki auto confirmation status, but we can

266 00:29:54.920 --> 00:30:01.060 Francisco Navas: give you test with the auto confirmation status. No problem. You just have to add your name to the list.

267 00:30:01.070 --> 00:30:02.350 Francisco Navas: The list

268 00:30:02.490 --> 00:30:04.769 Francisco Navas: is, where's the list

269 00:30:05.750 --> 00:30:08.030 Francisco Navas: should be at the top of the chat.

270 00:30:08.910 --> 00:30:10.600 Francisco Navas: and I'll repost it. Now.

271 00:30:13.640 --> 00:30:15.269 Francisco Navas: if you go in this, Doc.

272 00:30:15.780 --> 00:30:18.090 Francisco Navas: right below the

273 00:30:18.810 --> 00:30:21.179 Francisco Navas: agenda, there's a sign up

274 00:30:21.950 --> 00:30:23.310 Francisco Navas: to add effect.

275 00:30:23.700 --> 00:30:26.540 Francisco Navas: And then, actually, I will post right now

276 00:30:27.260 --> 00:30:30.030 Francisco Navas: the yeah URL for downloading

277 00:30:30.530 --> 00:30:31.939 Francisco Navas: in the chrome store.

278 00:30:33.360 --> 00:30:34.930 Francisco Navas: So you can go straight there.

279 00:30:38.220 --> 00:30:42.319 Francisco Navas: You're interested. You should be able to download it straight from that page.

280 00:30:43.850 --> 00:30:45.440 Francisco Navas: and yeah, leave us

281 00:30:45.620 --> 00:30:49.090 Francisco Navas: your name on your username, your end, Wiki username

282 00:30:49.150 --> 00:30:52.619 Francisco Navas: in that, Doc. If you want to give this a run

283 00:30:53.600 --> 00:30:56.309 Francisco Navas: or email, me, or whatever you know, reach us, however, you would like.

284 00:30:56.370 --> 00:31:01.730 Francisco Navas: And yeah, you can also leave feedback directly through a form which can be anonymous

285 00:31:02.120 --> 00:31:06.990 Francisco Navas: inside the extension itself. It says, feedback on the bottom next to login, so you can add that.

286 00:31:07.320 --> 00:31:09.160 Francisco Navas: And I think there's 1 more slide

287 00:31:09.180 --> 00:31:12.199 Francisco Navas: things for me to say, and I'll stop yapping at you.

288 00:31:14.150 --> 00:31:14.950 Francisco Navas: yeah, I can.

289 00:31:14.950 --> 00:31:33.830 Maryana Pinchuk: I can take this one. Thanks, Francisco. Yes. So we have this dev version available, and we're giving it to you. Our loyal future audiences, followers 1st before publishing it, publicizing it more widely next week. So I'll be in Wikipedia, in Poland.

290 00:31:34.220 --> 00:31:57.349 Maryana Pinchuk: and we'll have a future audiences talk on August 10.th So if anyone is coming to Wikmania, please come to my talk it will be on the last day, so I'm guessing attendance might start to fizzle but please come. And our plan is to make this available to any anyone who has auto confirmed rights on English Wikipedia, which I believe, just means

291 00:31:57.350 --> 00:32:05.210 Maryana Pinchuk: you have created an account. That is, it's older than 4 days, and I think it's you've made at least 10 edits. Correct me if I'm wrong.

292 00:32:05.210 --> 00:32:21.769 Maryana Pinchuk: Experts on auto confirmed status 10 edits, right? So you don't have to be an English Wikipedian like who's been around forever? If you're if you're home, Wiki is another project but if you do meet that bar on English Wikipedia, you'll be able to use the

293 00:32:21.770 --> 00:32:39.810 Maryana Pinchuk: production version, and we will be putting out a note on the English Wikipedia village pump to let people know that this is happening and this new version that we'll be demoing starting next week. Our intention is to make it post to English Wikipedia talk pages rather than to test Wiki.

294 00:32:39.810 --> 00:32:56.979 Maryana Pinchuk: This is to kind of test the full experience a little more, and to publicize it a little bit more broadly among the community of active editors. So in the talk Page post there will be a note with a link to the project, an explanation of what it is and an ask for more people to test.

295 00:32:57.368 --> 00:33:08.871 Maryana Pinchuk: So we're trying to see if this this, in its current state or something like it, could be valuable to to Wikipedians, who are already active, who are already looking for

296 00:33:09.240 --> 00:33:18.419 Maryana Pinchuk: sources outside of Wikipedia to bring in to Wikipedia, and also soliciting feedback on ways that this could change or improve.

297 00:33:18.490 --> 00:33:38.140 Maryana Pinchuk: So, for example, we know that some of the AI assessment language is maybe not quite perfect, gets a little murky, so we'd love your help in making sure that we're communicating the right things to users who are kind of going through this workflow.

298 00:33:38.450 --> 00:34:01.429 Maryana Pinchuk: Also sort of bigger thinking and suggestions around how this might look. If we were to open it up to more than just experienced Wikipedians, if we were to say, make this available to anyone in the world who wanted to add a fact. I think there are a lot of pros and cons to that. A lot of opportunities to to enable participation on Wikipedia to new audiences.

299 00:34:01.440 --> 00:34:19.230 Maryana Pinchuk: but also a lot of risks that we're very aware of around people abusing it, or contributing infowars or other unreliable sources to to our projects and creating a lot of work for the community to have to kind of clean up

300 00:34:19.300 --> 00:34:20.649 Maryana Pinchuk: and deal with.

301 00:34:20.659 --> 00:34:38.839 Maryana Pinchuk: So we really want to think through. What could a good experience look like? Could there be a way to moderate that type of content or triage it in some way. So it doesn't create a bunch of work. And with the power of AI, I think we can start to do more of that kind of synthesis

302 00:34:38.840 --> 00:34:52.679 Maryana Pinchuk: and analysis that we're bringing to show to editors who use this product. So thoughts around that are going to be really welcome. And yeah, I'm going to stop now and see

303 00:34:53.475 --> 00:34:54.130 Maryana Pinchuk: what

304 00:34:54.310 --> 00:35:04.189 Maryana Pinchuk: what the room is feeling. I've been kind of briefly scanning chat. So if anyone wants to be brave and unmute and speak to your impressions, ideas, thoughts.

305 00:35:04.510 --> 00:35:07.519 Maryana Pinchuk: Yes, Samuel, please. You have the floor.

306 00:35:08.467 --> 00:35:16.529 Samuel Breslow: Thanks so much for having this conversation and demoing this, I think these features look really cool.

307 00:35:16.980 --> 00:35:22.270 Samuel Breslow: I was curious just since you talked to the beginning about the future audience teams role

308 00:35:22.460 --> 00:35:26.219 Samuel Breslow: as doing these kind of quick experiments rather than

309 00:35:26.814 --> 00:35:29.149 Samuel Breslow: more built out products.

310 00:35:29.606 --> 00:35:37.519 Samuel Breslow: I'm I'm glad to see that since I think there's a tendency to have this kind of gulf between

311 00:35:37.620 --> 00:35:42.749 Samuel Breslow: things that are just like user, maintain scripts. And

312 00:35:42.780 --> 00:35:44.779 Samuel Breslow: so these very

313 00:35:44.970 --> 00:35:48.200 Samuel Breslow: can be janky type products

314 00:35:48.210 --> 00:35:57.090 Samuel Breslow: that can be made quickly on the one hand, and the features that the foundation makes, which are very well considered very cautious.

315 00:35:57.583 --> 00:36:00.686 Samuel Breslow: But tend to take a lot longer.

316 00:36:01.550 --> 00:36:02.600 Samuel Breslow: and

317 00:36:03.370 --> 00:36:06.050 Samuel Breslow: yeah, it's interesting to be able to have some

318 00:36:06.140 --> 00:36:10.669 Samuel Breslow: some things in a more middle ground. I'm curious whether there's any pathway

319 00:36:10.720 --> 00:36:19.500 Samuel Breslow: currently that you are considering, for if any of these experiments go particularly well, and there's interest in

320 00:36:19.780 --> 00:36:30.679 Samuel Breslow: having the feature be available beyond the scheduled end date. Would you hand them off to another team to like, build out more fully or

321 00:36:30.770 --> 00:36:33.450 Samuel Breslow: be able to keep them around in some way

322 00:36:33.860 --> 00:36:35.350 Samuel Breslow: that would.

323 00:36:36.030 --> 00:36:42.260 Samuel Breslow: you know, hopefully, not create maintenance work, but allow their benefits to continue.

324 00:36:42.600 --> 00:37:08.040 Maryana Pinchuk: Yeah, that's a really great question. Thank you. Yes. So the process right now kind of looks like the the latter what you outlined. So if we hit on something that's really successful. So let's say we put this out there, and every Wikipedian out there wants to use it, and not just English Wikipedians, but every Wikipedian and and Wikidata, and and lots of other folks in our movement are really excited about this

325 00:37:08.409 --> 00:37:28.740 Maryana Pinchuk: we would then come to the the kind of planning process that the foundation has which occurs every year sort of starts really in January, we really start thinking about what we're gonna prioritize for the next fiscal year and and goes into April to June to like, actually write the draft

326 00:37:29.010 --> 00:37:52.520 Maryana Pinchuk: and, you know, start to like collaborate with teams kind of pitching what they think is really important and gonna be impactful. We would come to that process and say, Hey, look! We're we're getting a ton of great feedback from our community. People really want this thing, and they want this thing for real. They don't just want, you know, a browser extension that's like only for one browser and one language. We really need like to to give this to a full

327 00:37:52.520 --> 00:38:19.640 Maryana Pinchuk: product team, and then that product team would take it and would build it in the right way, quote unquote right? So they would make it available to more users of more browsers and more languages, and you know but the idea would it would be one of many that are kind of going into that process and through internal discussion through discussion with the communities. So we publish a draft of our annual plan on Meta every year

328 00:38:19.640 --> 00:38:33.629 Maryana Pinchuk: and take feedback and do really try to incorporate community feedback as well to decide. Okay, like, yes, this is important. Lots of people want this. But there are 20 other things that are maybe even more important and impactful. And and going to really.

329 00:38:33.630 --> 00:38:58.029 Maryana Pinchuk: you know, move the needle of increasing editors or readers, or whatever. This is a collaborative discussion that we have to engage in with our movement. So yeah, so it's really just listening, getting insights and making recommendations in in the sort of normal annual planning process. But there are also kind of

330 00:38:58.170 --> 00:39:18.650 Maryana Pinchuk: other ways that this could show up, so we we might not see that every comedian in the world wants to use something like this. But we're really hoping to learn something valuable about the process. So there might be other kinds of recommendations that come out of it, such as and we've we've seen this kind of floated again and again. In various other contexts. But we're kind of hitting the same

331 00:39:18.973 --> 00:39:29.010 Maryana Pinchuk: issues while building this product is that we don't really have like a structured reference bank or a structured bank of claims, with references attached to them on on Wikipedia or

332 00:39:29.010 --> 00:39:42.599 Maryana Pinchuk: anywhere. It's Wikidata kind of it's it's it's sort of like that. But but not quite and this this idea has been floating around for a while in lots of different ways. You know, the Wiki site community has been really advocating for for that

333 00:39:42.890 --> 00:40:11.979 Maryana Pinchuk: and it may be the case that maybe that's something we'll learn a lot more about and and have something to to bring to an annual planning conversation around. So it could look it could look like. Yes, this is the product we've we found it. We found the the perfect product that's going to really solve a a major need. And we're gonna recommend that it's built out for real. Or we've learned some things by doing this that are touch on other parts of the whole product ecosystem. And we need to make recommendations about those

334 00:40:12.850 --> 00:40:14.900 Maryana Pinchuk: I see. Marshall has a hand.

335 00:40:15.000 --> 00:40:17.380 Maryana Pinchuk: Would you like to chime in Marshall.

336 00:40:17.780 --> 00:40:34.249 Marshall Miller: Yeah, thanks. I think I just wanna take this opportunity to like, point out a little bit more of the mechanics of how something like this works in our organization. So, for instance, I'm a director of product, and Mariana reports to me. And so I'm in constant contact with Mariana about this work and about these learnings

337 00:40:34.250 --> 00:41:02.339 Marshall Miller: and other teams that report to me include, like the editing team, the growth team, the web team. And these are some of those teams that are building those like longer term high scale features. And so, you know, it's through that like direct communication that the various product managers and teams involved in this is is how we're exchanging information. And and like, we've deliberately set up our org structure so that it's not like Mariana has to be off on the sidelines like insisting that she's got something valuable. It's like

338 00:41:02.610 --> 00:41:07.689 Marshall Miller: work structure, wise. She's plugged into the teams that would actually be doing that implementing.

339 00:41:07.980 --> 00:41:18.760 Marshall Miller: and in terms of a specific example, kind of like Mariana referenced at the beginning like working on the Chat Gpt. Plugin, we learned a lot about how Llms interact with Wikipedia content.

340 00:41:18.910 --> 00:41:42.779 Marshall Miller: And it gave us confidence that we can figure out how to make them interact productively. And so then for the year that we're starting now this fiscal year. There are teams that are working on features like this edit check feature that's going to attempt to use an Llm. To guide new editors to make better edits. And we're doing that because of the confidence we gained from the future audience. Experiment about the Chatgpt Plugin.

341 00:41:42.930 --> 00:41:47.950 Marshall Miller: So that's like it's not like we directly translated the feature. It's like we translated the learnings.

342 00:41:51.940 --> 00:41:52.960 Maryana Pinchuk: Thank you, Marshall.

343 00:41:54.390 --> 00:42:02.090 Maryana Pinchuk: Alright. Other questions feedback on add effect or Meta process questions also welcome

344 00:42:02.600 --> 00:42:03.890 Maryana Pinchuk: gasolana.

345 00:42:05.110 --> 00:42:15.155 Ilana Strauss: Yeah, I don't have anything super technical to say or anything. But I just think this is so cool, like, I don't know. I I, it's very cool to like see it alive.

346 00:42:15.740 --> 00:42:19.949 Ilana Strauss: I don't know. I could just I could see it having an impact on

347 00:42:20.080 --> 00:42:23.440 Ilana Strauss: kind of so many other users who would never

348 00:42:23.560 --> 00:42:25.440 Ilana Strauss: spend that much time.

349 00:42:25.500 --> 00:42:30.394 Ilana Strauss: you know, just all of a sudden being able to be so involved, so great job.

350 00:42:33.380 --> 00:42:45.200 Maryana Pinchuk: Thank you. And for for the benefit of the folks in the room. Do you wanna say a little bit more about how you came into the early kind of design process of this project and sort of where you're coming from in the kind of

351 00:42:45.810 --> 00:42:47.350 Maryana Pinchuk: Wiki work verse.

352 00:42:48.620 --> 00:42:51.339 Ilana Strauss: Oh, man, yeah. Well, I I mean just to

353 00:42:51.610 --> 00:42:56.470 Ilana Strauss: just total coincidence. We're all just like sitting on a train. Going to.

354 00:42:56.760 --> 00:43:03.809 Ilana Strauss: Their 2 are from like a conference and just started chatting about it. He was kind of organic.

355 00:43:04.470 --> 00:43:11.739 Ilana Strauss: But yeah, I run so a crowd sourcing, fact, checking organization. So sort of

356 00:43:12.260 --> 00:43:13.700 Ilana Strauss: I've described it as like

357 00:43:13.760 --> 00:43:29.799 Ilana Strauss: Wikipedia for fact checking to people. So yeah, I kind of came to a conference just to sort of meet people, and you know, see how everybody is thinking kind of in this crowdsourced information. Space.

358 00:43:39.920 --> 00:43:42.230 Maryana Pinchuk: Other questions. Thoughts.

359 00:43:45.785 --> 00:43:46.439 Maryana Pinchuk: Critique

360 00:43:47.350 --> 00:43:49.929 Maryana Pinchuk: Marshall, do you still have your hand up, or is this a legacy hand.

361 00:43:49.930 --> 00:43:55.339 Marshall Miller: It's a new hand in which I have a question to prompt the group when we run out of stuff. People already want to say.

362 00:43:55.590 --> 00:43:57.090 Maryana Pinchuk: Well, critique, please take it away.

363 00:43:59.300 --> 00:44:01.420 Pratik M: Marshall can go 1st if he likes.

364 00:44:01.930 --> 00:44:03.500 Marshall Miller: No, no, I think you should.

365 00:44:04.340 --> 00:44:20.488 Pratik M: Okay, I had a few thoughts. I am new to Vicki in general. I have obviously known about it for a long time, but I haven't edited or contributed in that sense. So I'm looking at it from that lens a little bit.

366 00:44:21.000 --> 00:44:36.980 Pratik M: I was thinking in terms of this Llm. In the loop as Llm. As an assistant kind of a way of looking at things. It could be really interesting if this tool could be a tutor of sort for the ways of

367 00:44:36.980 --> 00:45:00.750 Pratik M: Wikimedia, of you know what the rules are, how to learn about sources, how to learn about where things go, because adding a fact to an article is really contextual, like where something is missing. You need to have that information first.st If you're going to add something meaningful. It can't be just you find something online, and you think it might be useful over there. So I think

368 00:45:00.770 --> 00:45:10.150 Pratik M: that could be an interesting RAM to having more full-time contributors, or people who want to actually contribute properly long term.

369 00:45:10.160 --> 00:45:18.840 Pratik M: And from the other side of things I was also thinking it it might be interesting to have these facts come in. Not.

370 00:45:18.840 --> 00:45:42.259 Pratik M: I mean, obviously, this is a prototype. So the top page is just going. But it could be like a collection of sorts, like an arena channel, or like a pinterest board, or some something of that nature, where facts, instead of adding a fact, you're like adding a recommendation, or adding a direction or a thread of some sort from where things can go further.

371 00:45:42.410 --> 00:45:46.260 Pratik M: I think those are some of the thoughts I wanted to put forward.

372 00:45:50.610 --> 00:45:53.700 Maryana Pinchuk: Awesome. Thank you. Yeah. I think, the

373 00:45:53.980 --> 00:46:03.909 Maryana Pinchuk: the 1st point that you have about kind of using this as a way to educate newer contributors or people who are newer to the wikiverse is a really interesting one.

374 00:46:04.224 --> 00:46:28.720 Maryana Pinchuk: And it makes me think, you know, we've got this kind of very simple message about the the quality of the source that we show to the user. But there could be a lot more there around, not just not just saying like, Oh, sorry this. The source doesn't work on Wikipedia. But maybe bring in more of that like instruction. Learning about how Wikipedia does work, and who decides which sources are notable. And how

375 00:46:28.720 --> 00:46:30.610 Maryana Pinchuk: can contribute to that discussion? And

376 00:46:30.906 --> 00:46:45.739 Maryana Pinchuk: yeah, I think there are a lot of really interesting directions to explore, and that and then the the other point you made about you know where the facts could go is creating new ways of of collecting these contributions. I think. Also, yeah, very interesting. And

377 00:46:46.122 --> 00:47:07.609 Maryana Pinchuk: I'm curious. Once people get a chance to to play around with it a little more. That'll be one of our prompt questions in the feedback is like, Where where could these things go? To? Really be maximally useful? And you know, both for the contributor to make them feel like it went somewhere, but also for our communities to have a place to sift through this stuff and find the really good valuable contributions.

378 00:47:07.700 --> 00:47:09.019 Maryana Pinchuk: Yeah.

379 00:47:09.300 --> 00:47:13.230 Maryana Pinchuk: what is what is on the Pdf. Dare I ask?

380 00:47:15.850 --> 00:47:18.470 Sam Klein: Them into a scratch space, a scratch face for

381 00:47:19.540 --> 00:47:33.589 Sam Klein: notes and other things. I I feel right now for for a lot of test cases we're overloading top pages way that we did for notifications and template messages. And maybe we can do something.

382 00:47:34.130 --> 00:47:34.870 Maryana Pinchuk: Hmm.

383 00:47:35.250 --> 00:47:36.259 Sam Klein: Better than that.

384 00:47:36.970 --> 00:47:38.609 Maryana Pinchuk: Yeah, a really good point.

385 00:47:40.656 --> 00:47:41.123 Maryana Pinchuk: Yeah.

386 00:47:41.600 --> 00:47:43.504 Maryana Pinchuk: There's so many wikis

387 00:47:44.723 --> 00:47:58.589 Maryana Pinchuk: Marshall, did you? Or did anyone else have any thoughts around that topic. I I know it's a lot to think about like, where should this go? But I'm curious. If even in just looking at it, you you had some thoughts around that anyone in the room

388 00:48:01.290 --> 00:48:08.510 Maryana Pinchuk: does anyone feel like talk pages on English? Wikipedia would absolutely not be the place, even for an experiment.

389 00:48:10.050 --> 00:48:23.010 Samuel Breslow: i i i think I share the concern that others have brought up, that a lot of talk pages are very inactive, and that putting things there can make them go into a void. I think both of the ideas that you mentioned.

390 00:48:23.443 --> 00:48:48.550 Samuel Breslow: Putting them on project talk pages which, while still sometimes an active, or at least a little more active that could be good, and if there was a dedicated queue of it, that could also work really well. The question would be, how much of a backlog with that queue form which is probably a function of just how interesting is it to be a reviewer doing this task?

391 00:48:48.873 --> 00:48:51.000 Samuel Breslow: What we tend to see is that

392 00:48:51.170 --> 00:48:53.329 Samuel Breslow: for the fun type jobs

393 00:48:53.500 --> 00:49:10.453 Samuel Breslow: get done a lot or the easy jobs get done a lot, whereas ones that are challenging or unrewarding, like, you know, answering coi edit requests things like that tend to, or, you know, approving articles at Afc. Those can develop extremely long backlogs.

394 00:49:11.210 --> 00:49:12.310 Samuel Breslow: I'm

395 00:49:12.650 --> 00:49:24.899 Samuel Breslow: like, I I think this would probably be most analogous to just like the backlog of citation needed tags on claims which what we've seen is that like.

396 00:49:25.640 --> 00:49:27.979 Samuel Breslow: if there's attention paid to an article.

397 00:49:28.610 --> 00:49:51.899 Samuel Breslow: the citation needed, tags will get addressed. But there's not really interest among the editor community in going through the full backlog and just like actively searching out citation, needed tags in order to try to fix them. There are just so many of them, and there's a few tools that allow you to do that. But it's not something that most editors seem to want to spend their time doing.

398 00:49:53.560 --> 00:50:12.710 Marshall Miller: Samuel. One of the motivations behind. The idea of the putting it on the talk page is, I think, that we've built products in the past that created some new queue that people had to go through, and we were loath to add another queue to all of the queues that Wikipedians have to think about

399 00:50:13.090 --> 00:50:22.890 Marshall Miller: and it was sort of like, what existing queue can this get tacked onto? So it's not like another thing on the to do list like it would. Might come across people's watch lists naturally.

400 00:50:22.940 --> 00:50:34.030 Marshall Miller: So that's sort of like the motivation. And may maybe there's like a different, better queue that people look at as a matter of their like Wiki work, where something like this could could feed into.

401 00:50:36.434 --> 00:50:40.919 Liam Wyatt (WMF): If I could jump in. I actually can't find my hand out thing as well, Marshall. I don't know where.

402 00:50:40.920 --> 00:50:42.539 Marshall Miller: Under the react button.

403 00:50:44.270 --> 00:50:46.249 Liam Wyatt (WMF): Okay, I will

404 00:50:47.120 --> 00:50:55.900 Liam Wyatt (WMF): preemptively react by holding my hand up. To Marshall's Point. There, there have been previous products that created new tabs.

405 00:50:56.010 --> 00:51:08.499 Liam Wyatt (WMF): So I think, was the adequate feedback tool, had a edit talk, and then suggestions that came from the general public. Tab. But of course, creating a new tab is a new workflow, a new place. People haven't

406 00:51:09.310 --> 00:51:14.139 Liam Wyatt (WMF): worked in before and also doesn't have the ecosystem of tools and templates

407 00:51:14.660 --> 00:51:15.610 Liam Wyatt (WMF): board.

408 00:51:16.177 --> 00:51:19.780 Liam Wyatt (WMF): There has been a suggestion, as I put in the chat here, that these

409 00:51:19.950 --> 00:51:21.429 Liam Wyatt (WMF): citation needed

410 00:51:21.740 --> 00:51:28.929 Liam Wyatt (WMF): sorry add effect. Suggestions should go into a central to-do list, and then that could be translated to the respective talk pages.

411 00:51:29.150 --> 00:51:31.480 Liam Wyatt (WMF): And people could work through that to-do list.

412 00:51:31.590 --> 00:51:38.289 Liam Wyatt (WMF): Currently, it's designed around. That's an idea. Currently, it's designed around the idea that the suggestions go onto the relevant talk page.

413 00:51:38.940 --> 00:51:41.360 Liam Wyatt (WMF): and that's where they live, and people respond to them

414 00:51:41.440 --> 00:51:45.969 Liam Wyatt (WMF): on the basis of the responder is interested in the subject.

415 00:51:46.230 --> 00:51:48.759 Liam Wyatt (WMF): not interested in the idea of

416 00:51:48.990 --> 00:51:52.339 Liam Wyatt (WMF): off-site submissions as a worklist per se

417 00:51:53.368 --> 00:51:55.520 Liam Wyatt (WMF): they. So the people would be

418 00:51:55.970 --> 00:52:02.439 Liam Wyatt (WMF): Wikipedians would be interacting with the new suggestions on the top page of the article they care about.

419 00:52:02.530 --> 00:52:09.979 Liam Wyatt (WMF): rather than going to hunt for a list of new suggestions from off-site and working through them one by one

420 00:52:10.880 --> 00:52:15.410 Liam Wyatt (WMF): might be a valid use case. But this is the one we're testing now.

421 00:52:18.798 --> 00:52:22.830 Marshall Miller: Here's something I wanted to ask the group before we have to depart. So

422 00:52:23.300 --> 00:52:31.670 Marshall Miller: the team's kind of doing something counterintuitive here, we're trying to investigate future audiences like people that don't currently participate in Wikipedia.

423 00:52:31.890 --> 00:52:36.999 Marshall Miller: But we're building the add effect extension. And then we're gonna

424 00:52:37.050 --> 00:52:39.689 Marshall Miller: pitch it to existing Wikipedians.

425 00:52:39.700 --> 00:52:58.210 Marshall Miller: because we think that we can get people, those people to use it right. Existing Wikipedians will get the concept. They'll give it a try. But that's what we're hoping. And then from that kind of usage, and from those reactions that will help us learn like, could we imagine non-wikipedians picking something like this up and getting involved that way.

426 00:52:58.330 --> 00:53:03.179 Marshall Miller: and so given that we've got a bunch of Wikipedians in the room. My question is like.

427 00:53:03.310 --> 00:53:19.890 Marshall Miller: you know, concretely, the team intends to take this to English village pump and say, like, here's something we've been working on. People. Wanna try this out and so wanted to kind of take your temperature on like. How do you think that'll be received? Do you think English Wikipedians will want to do this, or what might you?

428 00:53:19.920 --> 00:53:25.590 Marshall Miller: What advice do you have on like how to explain it, or how to encourage that kind of usage? Or maybe what might they be worried about?

429 00:53:32.930 --> 00:53:39.180 Samuel Breslow: Yeah, I think any tool that is ultimately destined for

430 00:53:40.060 --> 00:53:44.520 Samuel Breslow: like people who are not currently Wikipedia editors is going to

431 00:53:45.120 --> 00:53:46.849 Samuel Breslow: raise some

432 00:53:47.040 --> 00:53:50.504 Samuel Breslow: hesitations just because of the history of

433 00:53:51.490 --> 00:53:56.337 Samuel Breslow: Was it like suggested edits, or something like in 2011 way far back?

434 00:53:57.030 --> 00:53:58.190 Maryana Pinchuk: Back. Tool, yes.

435 00:53:58.190 --> 00:54:05.077 Samuel Breslow: That's what it was. And so yeah, I'm sure you all have that like very much in mind.

436 00:54:05.880 --> 00:54:06.950 Samuel Breslow: and

437 00:54:08.090 --> 00:54:14.470 Samuel Breslow: yeah, I think, like for me. Personally, I find it always persuasive when there's comparisons of like

438 00:54:16.560 --> 00:54:27.489 Samuel Breslow: I forget which. Recent tool it was, but it was looking at like the revert rate between newcomers not using the tool and newcomers who are using the tool and

439 00:54:27.830 --> 00:54:33.260 Samuel Breslow: how it's a success for us when we have a lower revert rate for

440 00:54:34.585 --> 00:54:35.440 Samuel Breslow: editors

441 00:54:35.450 --> 00:54:44.049 Samuel Breslow: who are using a tool, even if newcomers are still in general reverted or making deleterious edits a fairly high rate.

442 00:54:45.590 --> 00:54:46.880 Samuel Breslow: and so.

443 00:54:48.490 --> 00:54:55.179 Samuel Breslow: yeah, I, I think Wikipedians know that there's a hugely pressing need to

444 00:54:55.600 --> 00:55:05.709 Samuel Breslow: bring in more editors and get more people contributing to the project, because there's just so much editing work to be done, and such a lack of editors to do it.

445 00:55:06.850 --> 00:55:09.300 Samuel Breslow: I think other Wikipedians

446 00:55:09.690 --> 00:55:15.190 Samuel Breslow: is there, and I try not to fall into this trap myself, but when it comes to specifics

447 00:55:15.630 --> 00:55:19.079 Samuel Breslow: trying to actually get newcomers involved, then it's like.

448 00:55:19.330 --> 00:55:22.050 Samuel Breslow: Oh, my gosh! We've got this lot of people coming in

449 00:55:22.290 --> 00:55:28.790 Samuel Breslow: doing not great edits. And of course they're not going to be great because they're newcomers and don't know everything yet.

450 00:55:29.492 --> 00:55:31.539 Samuel Breslow: And that can

451 00:55:32.000 --> 00:55:38.870 Samuel Breslow: be where there's then trepidation and the community not wanting tools to be turned on.

452 00:55:39.020 --> 00:55:40.050 Samuel Breslow: So

453 00:55:40.420 --> 00:55:43.679 Samuel Breslow: yeah, I would just emphasize the like. This is

454 00:55:44.030 --> 00:55:47.410 Samuel Breslow: something that it comes to fruition in a more finalized form could

455 00:55:47.430 --> 00:55:54.610 Samuel Breslow: help us address all the content gaps that we have and the need for there to be more editors.

456 00:55:58.630 --> 00:56:00.650 Liam Wyatt (WMF): Samuel. Yes, and and that

457 00:56:00.930 --> 00:56:07.870 Liam Wyatt (WMF): slowly, Sully, careful. That's not scary. One approach is exactly why it's going to test Wiki.

458 00:56:07.970 --> 00:56:10.829 Liam Wyatt (WMF): and only to auto confirmed users

459 00:56:10.930 --> 00:56:18.099 Liam Wyatt (WMF): initially, even though, as exactly as Marshall said, that's not the intended eventual destination for this tool.

460 00:56:18.250 --> 00:56:22.650 Liam Wyatt (WMF): but by getting the existing community to feel comfortable with its

461 00:56:23.380 --> 00:56:24.490 Liam Wyatt (WMF): systems.

462 00:56:24.500 --> 00:56:25.819 Liam Wyatt (WMF): will hopefully

463 00:56:26.241 --> 00:56:34.690 Liam Wyatt (WMF): make that transition to when we hopefully open it up to a wider audience. People will go. Okay, we understand this is being done, and they come careful.

464 00:56:35.190 --> 00:56:38.320 Liam Wyatt (WMF): gradual rate, and not

465 00:56:38.580 --> 00:56:55.269 Liam Wyatt (WMF): scare everyone, and to immediately shutting down experiments. The other aspect is, and this might be a question for the audience and request of the audience is precisely because existing Wikipedians are the 1st people who are going to be using it.

466 00:56:56.770 --> 00:57:04.280 Liam Wyatt (WMF): The 1st thing we've seen a couple of people do is turn it on Wikipedia itself. I write a sentence or a paragraph in Wikipedia.

467 00:57:04.430 --> 00:57:21.160 Liam Wyatt (WMF): and expect that the tool will then go onto the Internet and find a footnote somewhere out there that will help improve the improve the citations of the sentence already written in Wikipedia. That's not how this tool works that's looking at the telescope through the wrong end.

468 00:57:21.400 --> 00:57:25.259 Liam Wyatt (WMF): And we need to make sure people understand that. That's

469 00:57:26.440 --> 00:57:30.209 Liam Wyatt (WMF): you know. Thank you for trying, but that's not how it's intended to work.

470 00:57:30.220 --> 00:57:33.304 Liam Wyatt (WMF): And if you see people complaining about that

471 00:57:33.840 --> 00:57:34.880 Liam Wyatt (WMF): we need to

472 00:57:34.970 --> 00:57:38.549 Liam Wyatt (WMF): find better ways of reorienting the behavior.

473 00:57:40.350 --> 00:57:42.609 Liam Wyatt (WMF): because we expect that will happen

474 00:57:42.760 --> 00:57:48.660 Liam Wyatt (WMF): more with advanced Wikipedians than with the eventual intended audience.

475 00:57:52.380 --> 00:58:14.760 Maryana Pinchuk: Yeah, thanks. Thanks, Samuel and Liam. I think we are at time. But thank you all so much for participating in this conversation reminder that we do these monthly-ish. We had a little gap last couple of months, but we're trying to get back on schedule, so please follow us, and we post on the talk page of Meta

476 00:58:14.760 --> 00:58:27.740 Maryana Pinchuk: Meta. Future audiences. So that's how you can stay up to date with what we're doing for those of you at Wikimedia. I'll see you there, and for everyone else. I'll see you next month, and we'll let you know how this goes. So thank you.

477 00:58:28.140 --> 00:58:29.020 Maryana Pinchuk: Ciao.