Ronen Tamari 0:00 are hallucinated references that are just completely made up. 0:04 And in this better networked attribution, then you can actually hover over something and see if that thing exists. 0:14 Just that, better formats, wipes out a whole side of the attack. 0:20 And then I think, again, leaning on some of the expertise that has built up in the social media space around moderation and that side, I think that is more mature than what is in the scientific space right now. Speaker B 0:34 Okay, that's all I've got time for right now. 0:36 Thank you very much for the questions. Speaker C 0:58 Hello again. 1:06 Is this thing on? 1:07 Yeah. 1:08 OK. 1:09 Yeah. 1:09 So how many of you have actually heard about Sembol already? 1:12 Oh, OK. 1:12 Well, nice. 1:14 I have like an if-else of like how to describe this. 1:18 Great, cool. 1:19 So, yeah, Semble is our social knowledge network for research built on App Proto. 1:24 And we put this talk right after Matt and Rowan because we see Semble as a bridge between the modular research work they're working on and the App Proto space we're all in. 1:37 So, yeah. 1:40 How many of you have seen this slide? 1:42 Okay, good. 1:42 Because I show it a lot in every talk I do, so I guess I haven't seen you guys yet. 1:49 Yeah, Sembl is aimed at the problem of information overload. 1:53 And as researchers, yeah, how many of you feel this in your day-to-day work? 1:59 Yeah. 2:00 Yeah, we all feel this kind of overwhelm, too much information, too many papers to read. 2:07 Yeah, our capacity to make sense of all the information is far lagging behind our access to it. 2:12 We can download so many papers from archive. 2:14 We have open access. 2:15 But yeah, it doesn't help us answer the questions of what do we read first, who to trust, what's important, what's the big picture, how does this all even matter? 2:26 And it's not just an individual problem for like individual researchers. 2:30 It's actually breaking science as a whole. 2:33 The exponential growth in publications is breaking the peer review and journal curation systems that were traditionally the ways that science made sense of research, and it just isn't holding up anymore. 2:46 AI alone isn't a silver bullet. 2:49 When I talk about this in SF, I have to explain more like why AI isn't the only solution. 2:54 I think the crowd here is like, yeah, we like people and we like social media, so we know that we're on Blue Sky for a reason. 2:59 We like talking to people, and that's an important part of science. 3:02 So if AI isn't used wisely, it will actually make the problem worse. 3:07 It will enslopify research like we heard from Matt. 3:11 So we need people to be involved somehow. 3:12 We need to help this guy. 3:15 And our insight at Kosmic is that this fellow isn't alone, okay? 3:21 We're all— we're all this fellow. 3:24 We have strength in numbers, and by sharing the same problems, we can share solutions and actually work on them together. 3:31 So the insight here is that when each of us is reading at our desks or on our iPads, We are generating unique insights. 3:40 We're even choosing, like, what am I choosing to read first this morning is a valuable signal when it's aggregated, and especially with experts that really know their subject matter. 3:49 The problem is that most of these insights currently are lost. 3:53 Joel Chan, Matt's collaborator, calls this creative exhaust, which I really like as a term. 3:58 So we're not harnessing these signals that could help us make sense of information together. 4:06 This is what it looks like when we win. 4:08 We have something we like to call co-augmented reality. 4:12 We can aggregate our insights into living maps that help us navigate the ocean of content, kind of like how navigation apps help us navigate and choose which roads to take. 4:25 We don't have this yet, but we have some puzzle pieces. 4:30 Science social media is a great start. 4:32 We're going to hear a lot more about it today. 4:34 Lots of researchers are already sharing a ton of great insights. 4:38 You can open a feed like Paper Skyjust that Sophie's going to tell us about later, and you'll just get a lot of takes, what's happening, what's trending, things you wouldn't read in the paper between the lines. 4:50 So it's a great stream of commentary and recommendations, but at the same time, it lacks support for diverse knowledge types. 4:58 It's not really designed for research. 4:59 And Rohan was talking about this. 5:01 A screenshot of a paper annotated is not the same thing as an annotation that's actually stored on research infrastructure. 5:08 So this is good for humans. 5:09 Like, as a human, we can read this and be like, OK, I want to read this paper now. 5:13 But for machines, it's a lot harder, just making it really hard to get all that information out. 5:18 So we need better ways of storing knowledge. 5:21 There's also something we call the iceberg effect, which is that researchers on BlueSky or Twitter or other social media, they're not sharing most of what they're reading because it's not really a space that's designed for that. 5:31 Most people aren't that extrovert that they're just like, I'm reading this this morning, I'm reading that. 5:36 You'll do it in a place like Goodreads maybe, or like quieter social apps, but not on traditional social media. 5:43 On the other end of the spectrum, we have another puzzle piece, which is the modular research tools like we just heard about. 5:49 They're great for supporting all the diverse knowledge types we need with all the provenance and all the structure. 5:55 But there's questions about usability and bringing them to where people are at. 5:59 This is an example I really like. 6:01 Nanopublications. 6:02 How many of you have heard of that tool? 6:04 OK, cool. 6:05 So Nanopublications are great. 6:07 They do all these things. 6:08 You can really store any kind of knowledge on them as a little knowledge graph. 6:11 Here's actually a Nanopublication announcing a paper that I've read. 6:14 So I put a DOI there, and I put a comment, and I can nanopublish that. 6:17 And people are kind of incredulous in the beginning that, like, really, you could nano-publish what you read. 6:21 But yes, this is, like I said, it's valuable information. 6:25 And you can even license it, right, and get all the provenance. 6:29 But most people aren't going to do that. 6:30 You can't even delete it. 6:32 So you can only retract it. 6:33 So, like, it's forever on the scientific record if you publish what you read. 6:38 So yeah, so there's friction, right? 6:39 And we need to kind of bridge these two worlds. 6:44 The third puzzle piece is actually, yeah, just something I recently read this blog post, and it's a really great insight, and it kind of ties together. 6:52 There's also academic research about this. 6:56 It's about a question of interfaces, and there's this kind of debates in the knowledge tools research about different knowledge interfaces, and there's a distinction between gardens and streams. 7:09 So feeds are like streams. 7:11 They're where the fast stuff happens. 7:13 It's chronological. 7:14 It's like the new relevant high engagement stuff where things arrive. 7:20 And then there's the knowledge gardens, which are slower. 7:23 They're more spatial, cross-linked, longer duration. 7:26 They're where the knowledge compounds. 7:28 And they're both important, but currently most of our kind of social media experience is very like feed heavy, stream heavy. 7:36 So Sembl is an attempt to bridge— yeah, connect all these three worlds. 7:42 And let's show you a quick— oh, before the demo, yeah, the play on words, it's— we're really inspired by the idea of an ensemble, a diverse group contributing to a coordinated whole. 7:55 And these are the questions we're interested in Sembl addressing. 8:00 Who's reading what? 8:01 Reading broadly construed, watching, listening, anything on media. 8:05 What do you think of it? 8:06 How does it connect to other knowledge? 8:07 And just really make all that sharing and discovery easy, valuable, and delightful. 8:14 This is— oh, shoot. 8:19 Yeah, so this is just a quick— so you all seen Symbol. 8:22 I mean, I guess— OK. 8:25 Yeah, you log in with your Atmosphere account. 8:29 Yeah, just to give kind of the basics, Symbal is about organizing cards into collections, so you can create cards. 8:35 Cards are basically bookmarks. 8:37 You add them to collections. 8:40 And that's pretty much it for the basic activity. 8:42 You can create personal collections that only you can add to, or you can create open collections that anyone can add to. 8:50 And then— So that's the kind of basics of bookmarking on Sembel. 8:57 One of the other key features, though, and this is kind of getting more into research-specific stuff, is what we call the Sembel page. 9:02 So you know how journals and preprint servers like arXiv, for every paper they have like a landing page. 9:09 And the landing page there is aggregating context about the research. 9:12 It will tell you, for example, in arXiv they have all these tools. 9:17 What code is relevant to this paper? 9:19 What datasets? 9:21 What other related papers are there? 9:22 So it's giving you all this extra context. 9:25 What we're doing with the Sembel page, which is when I click on that orange arrow, the button there, I open the overlay in Sembel, and this is an app proto version of that landing page that works for any webpage, not just what Archive decided they want to support. 9:40 So we want to really kind of democratize that landing page and bring in all the context that we want. 9:46 So this is the Sembel page. 9:47 You can see the URL up here is sembel.so, and then there's like the URL structure. 9:51 So yeah, for any URL, this works. 9:54 And there's a lot of information here. 9:57 I'll just show you some quick highlights. 9:58 So you can click here and look at the different collections that this paper is in. 10:04 You can see mentions on BlueSky or other atmosphere apps like Leaflet or other tools. 10:12 And you can also do connections, which is a new feature I'm excited to show you about because we haven't really talked about this much yet. 10:19 And it's very related to Matt's work on the Discourse Graphs. 10:22 So connections is when I click up here on the Semmel page, I click connect. 10:27 I didn't show you that earlier. 10:28 But you get this kind of popup and you can— like bookmarks are one URL. 10:34 You just bookmark a URL. 10:35 Connections are connecting two URLs. 10:38 You can connect this paper with another paper, and you can choose the kind of relation. 10:42 So we have some of the discourse graph relations, but also other kind of research-relevant stuff, like supplemental materials, code, data. 10:51 One use case I'm kind of excited about is just related work, because as a researcher, I come across papers and I'm like, oh, you guys didn't cite me. 10:57 I want to add my paper here to the references section. 10:59 So now you can do that for any research you want. 11:05 Another thing you can do with connections that we're really excited about is kind of like community notes. 11:11 So it's adding context to papers. 11:13 So this is an interesting example that happened a few months ago. 11:17 This paper was making the rounds. 11:18 It's a paper about human talent and performance and how you become a specialist in different fields. 11:29 but there was also a lot of criticism about it on social media. 11:32 But the problem was that it was kind of hard. 11:33 They're not connected, right? 11:34 Like, science isn't going to show you, oh, this paper got taken down on social media. 11:40 What you can do with connections is you can add helpful connections. 11:43 So it's kind of like community notes. 11:44 I can highlight, hey, this take on Blue Sky, this post is really thoughtful and important. 11:48 And I want people that are looking at the science paper to be aware of this post on Blue Sky. 11:54 So you can do that now, too. 11:57 We also have a few discovery features. 11:59 So yeah, you can explore the feed chronologically. 12:04 You can also filter it. 12:05 So if you want to choose— we're doing media type filtering. 12:08 So if you're interested, let's say, just in research, you can filter your feed for that, or podcasts, or audio, or books, or whatever. 12:15 You can also filter according to different people that you follow. 12:18 So we have our own social graph on Sembull. 12:21 You can also follow collections and not just people. 12:24 And I'll say a little more about that later. 12:28 We have just open search queries. 12:30 You can do and combine those with filters. 12:34 And the kind of interesting thing where everything here is community curated, so the quality tends to be higher because it's kind of just a community curated knowledge base. 12:42 It's also a limitation when it's small. 12:44 If people aren't interested in something, it's not going to be on Sembly yet. 12:46 But as the platform grows, we hope it'll just get more coverage. 12:52 And yeah, on the Symbol page, you can also do similar cards. 12:54 A lot of people like to use that to just find stuff. 12:56 If you add something on Symbol, you'll see similar cards. 12:58 And then you can— oh, it's like rabbit holes that you start connecting other stuff to. 13:03 So you open one card, and then you find a bunch more. 13:07 Notifications is also kind of different in Symbol, and it's got a few interesting patterns. 13:11 So related to the collections following that I mentioned, you see what people are following in terms of your collections. 13:17 And that's a cool dynamic that you can start realizing what people are interested in of all the content you're sharing. 13:22 Like maybe people don't like my cat videos, but they do like my design thinking papers that I'm sharing. 13:30 And then another kind of connection is the one here on the bottom where when people connect the cards in your library, you also get a notification. 13:39 So it's kind of like our— we don't have replies on Symbol, but This is a kind of reply, so it's riffing on your stuff, and I'm connecting it to something else. 13:47 And it's just really interesting to see how people are connecting things that you added to something that you were completely unaware of. 13:55 And yeah, maybe the thing most exciting to me, at least about Symbol, is just the @proto community. 14:01 I mean, and I can say it to you guys in person. 14:03 It's awesome. 14:04 Thank you. 14:05 It's been really great building in this in this community, just people are such hackers and coming at it in good faith and trying out new stuff and tinkering with it. 14:16 We have docs. 14:18 They're friendly for AIs and for humans. 14:21 And a few of the projects that we've seen— so there's one pattern that we call living libraries. 14:25 So whenever your website has a list of things that you're maintaining, you can just put the list on Symbol and show the interface on the website as usual. 14:35 So at ProtoScience, our project list is on Assemble. 14:38 It's back— like, the backend is Assemble list. 14:40 And as a kind of website maintainer, it's actually really useful because I don't need to mess around with HTML anymore. 14:44 I just add the thing to my Assemble list, and it appears on the website. 14:49 Nick Vincent, who's here in the back— maybe he'll tell us more about it at some point— but is doing a really cool library, personal library kind of thing where just curating a bunch of really good material about a certain field of research that he's an expert in. 15:01 And you have this kind of nice front end, and you can just navigate the different bibliography there. 15:07 And it's living, because they're always adding to it, and you can follow it. 15:13 Yeah, another really interesting thing of building an app proto was this interop, the interop that happens. 15:18 So when we were thinking about Symbol before we launched in November, we were thinking about browser extensions. 15:26 And we want to have people annotate stuff and highlight stuff. 15:28 And we just didn't get around to it. 15:30 It takes a while to build a good browser extension. 15:31 And then Margin came along. 15:33 And how many of you know about Margin? 15:36 OK, cool. 15:37 So Margin is an annotation tool, a browser extension on App Proto. 15:42 And we just talked to each other and decided to interop. 15:44 And so Margin shows Symbol activity, and we show Margin activity. 15:50 And we don't need to build a browser extension right now, because people can use Margin and kind of benefit from the best of both tools. 15:56 So it's this kind of interesting competition dynamic, and we're excited by it. 16:04 Another fun thing that's happening recently is the AI agents are coming, and they're using our annotation tools. 16:10 And this is a really interesting dynamic where— yeah, this is one on margin. 16:16 So yeah, it's doing highlights. 16:18 And this is using it on symbol, just adding things to libraries. 16:22 I know some people have built bots that they send a link to on Signal and tell it annotate this, and then it goes and creates these annotations. 16:31 It's pretty wild. 16:33 But yeah, I mean, we think it's going to be a really interesting use case where some of the AI slop is actually going to really benefit from having humans in tight loops. 16:41 So I think this is actually a really interesting pattern to watch. 16:44 And by having upvotes and downvotes, perhaps we can kind of better control quality and signal to these AIs what we want them to do. 16:54 Just some stats. 16:56 Yeah, we launched in November. 16:59 We have about 760 users, and this is just an interesting pie chart of the kinds of content people are sharing. 17:07 Research is actually only still a small part. 17:09 A lot of App Proto nerds that are just sharing every blog post that comes out about App Proto. 17:13 Which is great, because we have collections of 200 App Proto apps, and I never knew there were so many. 17:18 And there's those people that are like, collect them all. 17:20 And it's great. 17:21 So yeah, lots of knowledge here. 17:26 Yeah, this is one of my favorite memes. 17:28 I think this is kind of— you can drop this in basically for any challenge we're facing today as individuals or a society, and it works. 17:36 Information overload is going to eat us individual researchers unless we figure out our stuff and coordinate our sensemaking on App Proto. 17:43 Don't panic. 17:44 Organize or ensemble. Speaker B 17:47 Thank you. Speaker C 17:56 Maybe time for one or two questions. Speaker D 18:05 Great talk. 18:05 So I had one question about just like how you're saving links. 18:11 Is there a way that you're thinking about archiving just like the state of the website? 18:16 Because that's one thing that I've been using with Obsidian with their Web Clipper where I have an archive of it so that you can then look to see like how it changed or if the site dies. 18:27 So what are your thoughts on that? Speaker C 18:28 Yeah, it's definitely something that's on our roadmap. 18:31 There's a few different archiving services that would actually be interesting to use. 18:34 And yeah, we want to add it. 18:35 It would also improve the semantic search as well when you have the full text of the website. 18:43 Any other questions? Speaker B 18:44 Yeah. 18:45 There's something I've always meant to ask, but I've never gotten around to it. Speaker C 18:51 Oh, thanks. Speaker B 18:52 When you have context, like you mentioned, a little bit like Community Notes, is that auto-created? 18:58 [Speaker:AUDIENCE] Look for every post on Bluesky about that URL and display it, or is it only manually curated? Speaker C 19:03 [Speaker:CHRIS MESSIANA] The Mentions tab is automatic. 19:06 So we're just using the Bluesky search API. 19:11 But the connections are going to be human. 19:14 A human is going to create those manually. 19:15 So you can take a specific Bluesky post and say, I want to make that connection more salient. 19:26 Cool. 19:26 Awesome. 19:26 Thank you.