Maria Antoniak 0:00 Sound good? 0:00 Does that— that sounds good. 0:02 Okay, thank you so much. 0:03 All right, hello everyone. 0:04 I'm Maria Antoniak. 0:06 So happy to be here today. 0:07 I'm so excited to be here for Atmosphere Conf and for AT Science. 0:12 The energy is so positive. 0:15 Everything in the world is so bleak. 0:18 In the world, in science, in AI, in tech, on the internet, on social media, but Blue Sky and @proto and all of you people building things give me hope. 0:33 I am a computer scientist. 0:35 I'm an assistant professor at the University of Colorado Boulder. 0:39 I spent a lot of time in industry, kind of going back and forth between academia and industry over the last decade plus. 0:47 Relevant to all of you probably, I've spent a lot of time working on social media and studying online communities both in my research and in my industry work. 0:56 So I've spent time, for example, at Twitter Cortex, pre-Elon Musk, working on community notes, now known as community notes, but what was then known as Birdwatch. 1:07 Spent some time working at Facebook Core Data Science on groups. 1:11 Some other places including the Allen Institute for AI's Semantic Scholar team, it's kind of like Google Scholar. 1:19 In Seattle, some other places, and my research is in natural language processing and cultural analytics, so it's not really what I'm gonna talk about today, like my more NLP AI research, but I would just shout out, if any of you are having challenges in the stuff that you're building and you're using machine learning for something or NLP techniques, and if I can be helpful to you, let me know, or if I can connect you with someone who will be helpful, let me know. 1:43 My research intersects different fields like AI and the humanities, science of science, narrative medicine, online communities. 1:50 Again, won't be talking about a lot of that, but the science of science stuff might also be interesting to some of the people in the room, and in particular how AI is affecting scientific infrastructures and practices such as the archive. 2:02 And generally, I'm just really interested in using computational methods to study language, society, and culture, and social media is a big part of that. 2:11 And I'm also a social media user. 2:12 I'm a poster, I like posting and making making friends on the internet and then meeting those friends in real life, or in the physical world. 2:20 Online is real life. 2:22 And I'm really passionate about the open and free internet and protecting that space as both a space for privacy and for community. 2:31 We need both of these things enabled. 2:33 And for creativity. 2:35 And what I'm gonna talk about today is Leah. 2:38 It's a project that I have been working on with my social media friend Eugene Venitsky. 2:46 So Eugene is a professor at NYU, and together we've been trying to fill a void that's been left by the fall of academic Twitter. 2:56 Academic Twitter was a great place, an imperfect but a really great place to exchange ideas, meet other researchers, build knowledge, sensemake, et cetera, et cetera. 3:04 So now researchers are scattered, there are a lot of people on LinkedIn, There are a lot of people still on X, especially in AI, so I think anyone with kind of a more industry overlap is, a lot of them are still on X, but I am not and Eugene is not, so we've been building Leah, which is a researcher-focused social app on App Proto. 3:23 It includes a bunch of different features that I will show you, including verification, clustered paper discussions, safety alerts, stronger social signals, enhanced profiles, and community notes style annotation. 3:35 And we've been really inspired by the stuff that you all are building and other people on the protocol are building, projects like Paper Skygest that we'll, I think, hear more about later today, as well as Semble, Black Sky, Eurosky, Anasota, and all these other creative and community-focused apps that people are building. 3:55 Origin story, so yeah, Eugene and I are both really active on research social media. 3:59 We've both been really frustrated by our colleagues' reluctance to leave X. 4:04 And Eugene has faced some massive bullying on, like, I really would say bullying, like sometimes very toxic and personal bullying for his posts about AI, and he's not alone in that. 4:15 And then over our winter breaks in late December and early January, we were both on break, we're both professors, we both got kind of obsessed with Claude Code, and we had enough energy and frustration built up to attempt something to fix this bleak social media landscape. 4:30 And yeah, fun fact, we've not ever met in the physical world, and we had never actually spoken, despite building everything that you're gonna see today together via social media and Slack exchanges, until Ronan hosted a Zoom call in which we actually talked to each other. 4:46 So social media is actually good sometimes. 4:49 You can build great collaborations and meet really wonderful people via social media. 4:54 It's not all bad. 4:56 Our goals with Leah. 4:57 So first, one thing we're really frustrated with is, Blue Sky has so many good features. 5:02 A lot of the things people were complaining about were already built in, like moderation features, custom feeds, and we just wanted to make all that easier to discover, surface that to people who were new, who were coming from X or other platforms. 5:16 We also wanted to add new features, so I'll show you some of these. 5:21 Also fun things, just like we can add an edit, we can do whatever we want. 5:24 We can add an edit button. 5:26 Eugene added LaTeX formatting. 5:30 We're free, we can do it. 5:32 Highlighting papers and research content, so really building features for research, and also adding context and social signal. 5:40 So how is this person connected to me? 5:42 Are they a researcher? 5:43 Who do they collaborate with? 5:45 What is their affiliation? 5:46 All the kind of social signal that you might build up in person at a conference, getting to know someone's work, try to show that on LEAH. 5:54 Here's a little mood board from my Arena board. 5:58 Some pictures and quotes and stuff that have been inspiring me. 6:02 So I've also been thinking a lot about gardens and gardening and nurturing and like community co-op gardens, and I don't know, that's kind of what's in the background in my mind as I've been working on this project. 6:13 Green and growing things. 6:14 I'm really inspired by Lori Emerson's work. 6:17 She's a researcher also at the University of Colorado Boulder. 6:20 She has this new book called Other Networks that is gorgeous. 6:23 It's like this big hardcover, gold embossed, beautiful book about other ways of thinking about and other histories of social connection. 6:34 And I feel like everyone in the room would like love it, so you should find a copy of that book and you should also talk to Lori. 6:40 And then I love this little thread that Lori posted, I'll just read the end of it. 6:44 So she said, it reminds me of Latour's encouragement for us to become not the one who debunks but the one who assembles. 6:51 The critic is not the one who lifts the rugs from under the feet of the naive believers, but the one who offers the participants arenas in which to gather. 6:58 So we can complain about X all day, and I'm happy to do that too, but we also should build stuff for people and give them a place to go. 7:09 So, that's right. 7:13 I'm gonna— now, that was a lot of text, so now I'm gonna walk you through what we've been building on Leah, which is currently invite-only. 7:21 And the hardest piece of this and the piece I I've been thinking a lot about is verification. 7:24 This is the part that's kind of unbuilt right now. 7:27 Started out with a labeler and an Ozone service, then built our own UI instead of Ozone. 7:32 That was really easy and great to set up. 7:36 Users could apply for verification through a guided flow, and then moderators could triage and approve those requests, and also mass approve from lists, but Eugene and I have both gone back and forth a lot on whether to even have verification, how to verify, how to implement. 7:51 It feels like a big decision. 7:53 I think some things that we want are like composability, so like layers of verification. 7:58 So ORCID authentication might be one layer and having a.edu email address might be another layer. 8:05 And then maybe you can add these things together and we show those maybe as labels on your profile even if you're on Bluesky. 8:12 And then others again can like have some more context and social signal about who is this person. 8:18 Transparency, so seeing all of this. 8:21 Progressive onboarding, so don't overwhelm people. 8:23 You don't have to do any of this at first if you don't want, or maybe you just start with ORCID or with the.edu email address and then add more over time as you feel ready. 8:32 And then of course decentralization, so making this all available to other people who are building stuff. 8:40 Yeah, some ideas of like things that I think we might be able to build in here into these like levels of verification, but this is kind of brainstorming at this point. 8:47 And this is something that I would like, would really like to talk to more people about because this is the thing I'm working on in theory right now and would like to get right. 8:58 Here's what our moderation interface looks like right now. 9:01 So we have some built-in connections to ORCID to make this a little bit easier and have this nice flow. 9:09 So if I'm looking at the profile of someone I don't know personally, I can quickly see their OpenAlec profile, papers they've written, their affiliations, and understand whether this person really is who they say they are and that they're a researcher and I should verify them. 9:28 So we have already verified 472 people manually. 9:32 And then this is what that looks like on BlueSky and on Leah. 9:35 So on BlueSky, we've got a label. 9:37 It's just that little green verified researcher under Luke's name. 9:41 And on the right is the green checkmark by Vagrant's Photo. 9:47 So that's verification, but now let's walk through some of the more, the stuff that's a little more built out and fun to look at. 9:54 So logging in. 9:55 Here's Leah. 9:56 This is the landing page right now. 9:59 We have OAuth, and in theory it works with custom PDS as well, which is great. 10:07 This has been a little tricky and buggy. 10:09 It works and it doesn't work, but I think it's all been working for the most part lately. 10:15 And when you log in, this is the home screen. 10:17 I'm really proud of our tabbed feeds on the right. 10:20 I worked really hard on that. 10:23 And I'm gonna walk through a lot of what you're seeing on screen feature by feature. 10:27 So first, moderation and safety. 10:29 This is also a part of Leah that I've worked on a lot. 10:32 So we have this, if you click on moderation on the left, this is the page that will launch. 10:36 This was my page today. 10:38 So we have some safety alerts. 10:39 So Ted, who has over 20,000 followers— so someone with a ton of followers just quoted your post. 10:47 Little alert. 10:47 And you can click it away if you want. 10:49 You can set if you want these kind of alerts or not, like how many followers at which this alert should appear. 10:55 There's a little notification like, hey, you might be getting a lot of attention very soon. 10:59 You might want to tune in and maybe shut down your post now if it's something sensitive. 11:04 or not. 11:04 I wasn't worried in this case. 11:07 And then we have this kind of safety and harassment help center. 11:10 So hopefully like kind of a one-stop shop where if you are suddenly getting piled on, you could come to this page and go through these steps to feel safer. 11:21 And in particular, we have some extra blocking features. 11:24 This is inspired directly by the Block Party app, which was quite active on Twitter pre-Elon, and I believe for some other platforms, but I keep waiting for them to build something for @proto and I don't know what their plans are. 11:37 So you can, for example, I can look up a particular post and I can find all the accounts who reposted or liked that post. 11:45 Sorry, Sophie. 11:47 And you can block all of them and we include some signal here, like be careful, these actually, these accounts have a relationship with you, you might not wanna block these particular accounts. 11:57 This is really good for bad situations, like really bad situations where you're really getting piled on. 12:03 Notifications, we have this kind of extreme notification interface that's basically like Maria's personal notification interface and probably should get simplified, but it's like all this stuff that I wanna see. 12:17 And so we have a kind of typical set of notifications, your messages, my safety alerts again. 12:24 And then we also have some kind of triaging of new followers. 12:28 So this is also helpful for people with bigger accounts who are getting a lot of followers each day. 12:32 We show sections of people you may know, so people who have connections to people you're connected to, everyone else, and then suspected potential spam, very low follower accounts, or we look for some other signals. 12:50 Profiles have a lot of extra features. 12:53 Again, focused on research and academic affiliation, where you publish. 12:58 You can add some papers, your favorite papers that you've written recently. 13:03 We're pulling from, so if you include your ORCID, then we can connect to OpenAlex and we can pull out, for example, your co-authors. 13:10 And if your co-authors are also, if we also have their ORCIDs, then we can even link to their profiles on Leah as well. 13:16 So again, adding social signal. 13:19 And then here's Eugene's page. 13:21 So one thing you may have noticed, we have these colors. 13:24 So here there's a purple color around Eugene's profile picture. 13:28 We also have yellow and blue depending on whether you are mutuals, they're following you, or you're following them. 13:34 This is actually one of my favorite features on Lia and the one that I probably miss the most when I'm back on Bluesky. 13:39 Immediate social signal of, wait, am I following them or are they following me? 13:44 And we have this section of us, so our interactions between the two of us on every profile. 13:51 And then papers. 13:52 So this is where Eugene has been working a lot. 13:55 So each paper has a dedicated page. 13:59 So Eugene has been ingesting the full firehose, looking for paper URLs, creating paper pages for them, and then you can see all the discussion, quotes, replies, and other discussion about those papers. 14:13 And then you can— we also show then because we're pulling the full firehose, we have these discover panels where you can see what's trending. 14:19 So this was this morning, a bunch of people were talking about this science paper about sycophantic AI. 14:26 We also have articles, some more popular writing, and some other stuff. 14:32 And then again, each of these has a dedicated page where you can see all the discussion across the network. 14:39 Feeds. 14:41 Also, I'm starting to go a little bit over. 14:43 Is that okay? 14:44 Okay. 14:45 Feel free to cut me off. 14:47 So the Remix feed, this is a special feed that we have on Lia. 14:51 This is also kind of like, yeah, Maria's personal thing that I wanted. 14:55 But I think, but actually, I think someone actually, someone else requested this and then I customized it. 15:01 So whatever feeds, These are my feeds that I like, that I have pinned all the time. 15:07 You can choose which ones are included in the remix feed and how much, what proportion of posts, and then it all gets remixed into this remix feed. 15:15 And this is actually where I spend most of my time now. 15:19 We also have some— again, we're trying to surface these custom feeds and default feeds as defaults for our users, so feeds like Paper Sky Jest or our verified researchers feed, we just wanna make those defaults for all of our users so they don't even need to learn about what custom feeds are, they're already there for them. 15:40 And again, so yeah, we have these colors, people I follow, mutuals, followers, you can choose whether or not these appear, but I really like them. 15:48 Don't remember what I was trying to show there. 15:52 Community notes works roughly the same way as it does on X. 15:58 You can add a note, other people can see it, and other people on Leah can also rate the note up or down. 16:08 Looking forward, so our immediate goals, integrate with other services like Sembull, Graze, Leaflet. 16:14 For example, we have a bookmark feature that's fine, but why? 16:18 We should just have Sembull. 16:22 Integrate and create research lexicons with Sembull, Chive, and other people building academic apps. 16:27 Finalize our verification process, learn and integrate tools from BlackSky. 16:32 Their announcements this week were super exciting, and the moderation interface they were showing, I was like, yeah, let's adopt that. 16:38 Extend the community notes to be more full-featured, and then I think a big thing is supporting conferences and events. 16:45 Challenges, and this is the last slide. 16:48 Eugene and I are burnt-out junior profs who are teaching. 16:51 We haven't touched the code in a month, I think, at least. 16:53 We're also not at Proto experts. 16:55 I know I use some words wrong and I don't have the deepest technical foundation because I don't have time to read everything. 17:02 Also, everything you've seen is Claude. 17:06 We are both computer scientists and great coders, but this was also partly an experiment of how far can you get with Claude, and it turns out really far. 17:16 I've touched the code, I think, 3 times. 17:18 I do look at the code. 17:20 We need funding and we need time. 17:22 We need a backend engineer with security expertise so badly, and we so badly need a designer with a creative and beautiful vision. 17:29 So if you are those people, or if you can connect us to those people, or if you can connect us to money— time seems more impossible, but if you know how to get more time, let me know. 17:40 Help us. 17:40 Help us build Leah and, yeah, help us build a better ecosystem. 17:52 I went over, so we can do one question. Speaker B 17:55 And I'm also going to ask if you're in the audience and you have a seat like in the middle, maybe scooch in because we do have people coming in and out and we do have like full attendance. 18:04 So just make room for people coming in. 18:07 So one question. 18:08 Who's it going to be? Speaker C 18:15 Wonderful, wonderful talk. 18:17 Thank you so much. 18:17 I'm thinking a lot about the blurred lines between traditional researchers and citizen scientists of the future. 18:24 And so how are you approaching this? Maria Antoniak 18:27 Yeah, and that's important for verification as well. 18:31 We've also thought about journalists and how they could very well have an interest in joining a platform like VIA and be tuned into these conversations. 18:41 So I didn't really talk about this for verification, but we also had categories for like organizations and publication venues. 18:47 Maybe we'd also want categories for journalists, citizen scientists. 18:52 But yeah, open question, I would be very interested to hear other thoughts about this and like how gated or not Leah the app should be. 19:03 We've been thinking no gate, but then we have again this like social signal and like verification built in, but then like anyone could use Anyone could log on to Leah and see those conversations.