An inside look at X’s Community Notes | Keith Coleman (VP of Product) and Jay Baxter (ML Lead)

0:00 The work that you guys do has had such a tremendous impact on the way the world works. I wanna start with just giving people a brief understanding of what is community notes. Someone on X can see a post. If they think it's misleading, they can propose a note that they think other people might find informative. Other people can then Rate that note. We actually look for agreement from people who have disagreed in the past. And and what we see is when people actually have that sort of surprising agreement, that's what makes the notes so neutral and and accurate and well written, really, overall. There's many people that are very polarized. How do you deal with people that are like super anti-vax, super jam six? One philosophical thing that's important is that we want all of humanity to participate. And sometimes people are surprised by that. We have all of humanity. We then have the data to understand what notes will be helpful to actual humanity. Every post is eligible for notes. We shouldn't exempt Elon. We shouldn't exempt government figures. We should like everyone even advertisers can get notes. There have than external studies, you know, run by people totally independent of us who have found that if you take a post with or without a community note, that actually people's agreement with the core claims in the post does change.

1:11 If they see it with the note versus without. Is there anything else along the lines of just working for Elon within an org Elon runs that may surprise people? If I were to start a company in that company, it would be even leaner than I would have made it before. I've been amazed with just how much the team is able to accomplish. With a small group and I think because of a small group. Today, my guests are Keith Coleman, product lead for Community Notes, and Jay Baxter, founding ML engineer and researcher for Community Notes. This conversation may be my newest favorite podcast episode so far. Community notes is one of the most impactful.

1:48 and clever and also underappreciated products in the world right now. If you ever use X slash Twitter and you see a note underneath a tweet correcting the misinformation in that tweet, That is community notes. I've never heard a deep dive into the story behind the product and the team that built it, and I'm excited to bring you just that. We get into the surprising origin story of the product, how the algorithm actually works, how the algorithm emerged out of an internal contest within Twitter, the principles behind community notes, and why staying true to them has been so key to its success, also how it survived four different leaders, including Elon and Jack.

2:24 and why it's now a big part of the solution to solving misinformation on the internet. including recently being adopted by Meta as their main fact checking tool. This is an incredibly special episode, and I'm so excited to bring it to you. If you enjoy this podcast, don't forget to subscribe and follow it in your favorite podcasting app or YouTube. Also, if you become a subscriber of my newsletter, you now get a year free. of Notion and Superhuman and granola and linear and perplexity pro. Check that out at lenny's newsletter.com.

2:51 With that, I bring you Keith Coleman and Jay Baxter. This episode is brought to you by Work OS. If you're building a SaaS app. At some point, your customers will start asking for enterprise features, like SAML authentication and skim provisioning. That's where WorkOS comes in, making it fast and painless to add enterprise features to your app. Their APIs are easy to understand so that you can ship quickly and get back to building other features. Today,

3:19 Hundreds of companies are already powered by WorkOS. including ones you probably know, like Vercell, Webflow, and Loom. WorkOS also recently acquired Warrant. The fine grain authorization service. Warren's product is based on a groundbreaking authorization system called Zanzibar, which was originally designed for Google to power Google Docs and YouTube.

3:42 This enables fast authorization checks at enormous scale, while maintaining a flexible model that can be adapted to even the most complex use cases. If you're currently looking to build role-based access control or other enterprise features like single sign-on, skim, or user management. You should consider work OS. It's a drop-in replacement for Ots Zero and supports up to one million monthly active users for free. Check it out at workos.com to learn more. That's work os dot com.

4:15 This episode is brought to you by Product Board, the leading product management platform for the enterprise. For over 10 years, Product Board has helped customer centric organizations like Zoom, Salesforce, and Autodesk build the right products faster. And as an end-to-end platform, Product Board seamlessly supports all stages of the product development lifecycle, from gathering customer insights, to planning a roadmap, to lining stakeholders, to earning customer buy-in, all with a single source of truth. And now. Product leaders can get even more visibility into customer needs with Product Board Pulse, a new voice of customer solution. Built-in intelligence helps you analyze trends across all of your feedback and then dive deeper by asking AI your follow-up questions. See how Product Board can help your team deliver higher impact products that solve real customer needs.

5:03 and advance your business goals. For a special offer and free 15 day trial, visit productboard.com slash Lenny. That's product board. Dot com slash L E N N Y. Mm-

5:19 Keith and Jay, thank you so much for being here. Welcome to the podcast. It's great to be here. Thanks for it. Thanks for having us on. It's so my pleasure. I'm so thrilled to be having this conversation. The work that you guys do has had such a tremendous impact. And the way the world works. So many product teams are always talking about driving impact. I want to drive.

5:39 Like you guys have actually build things that have changed the world in meaningful ways and continue to do that. And I've never really heard the backstory of how community nuts came to be and how it works and all these things. So I'm really appreciative of you guys making time. To chat.

5:52 Yeah, first you know, thanks for saying that. That's that's why we built this thing. Uh is to help people and it it's great to hear it. It's great to see people enjoying it and Finding it useful. I wanna start with just giving people uh a brief understanding of what is community notes. I think a lot of people may have kind of heard about it.

6:10 kind of maybe see it on X as they scroll through. They see these notes, but they're like I don't actually know what this is. So can you just kind of briefly describe what is community notes? Community notes is a way for the people, like the public. to add context to posts that might be misleading.

6:26 The basic way it works is that uh someone on X can see a post. If they think it's misleading, they can propose a note that they think other people might find informative. Other people can then rate that note. And if the note is found helpful by people who normally disagree with each other, indicating that it's probably accurate, it's probably really neutrally worded, it's probably informative. Then it will show to everyone on X.

6:50 And the goal is just to get people more information about what they're seeing so they can make Better decisions than I was. Amazing. And I think like Hearing this, it's like Absurd that this works. I think when people originally heard this idea.

7:05 Like no way this is gonna work. And so just uh Dive a little bit deeper. Can you give us a sense uh a l a deeper understanding of how it actually works? Because I think it's the algorithm that you guys designed. That is so clever that allowed this to work. So

7:19 Talk a little bit about that algorithm. Yeah, so so I think a key uh misunderstanding a lot of people have, if they haven't really dived into details, is they kinda just think that maybe someone can write a note and it appears immediately, or Or we're just taking a majority rules vote uh of who thin the note's good. I think both of those approaches would probably lead to biased or inaccurate notes. I think the key thing uh really that we do is

7:43 We actually look for agreement from people who have disagreed in the past. Uh and and what we see is when people actually have that sort of surprising agreement. That's that's what makes the notes Uh so neutral and and accurate and well Written really overall.

7:59 Um, is just that people who are very polarized, um uh overall uh often can't find agreement when things aren't accurate, right? I I I think it also provides some good anti manipulation properties. I think people are often you know, it if you said uh I think like back in twenty twenty before we started building anything here, whether this could work at all. I I think a room of ML engineers would say, Oh, you have to keep it closed source. You know, people are gonna be manipulating this all the time. You have to use ground truth labels from fact checkers. There's no way That you could Like bootstrap the system without external labels, uh, but it's it turns out that you can do that. Um with with this kind of

8:38 Bridging based agreement algorithm is what we call it. Okay, so just to summarize and make it super clear, it's basically people someone writes a note. This information is false. What's like a good example just as we talk about this, like a classic example? A really really classic example is an AI generated image. Or an out of context image, like, look what's happening here, but it's actually from like five years ago in a different country on a different topic. Oh man, I've seen those so many times. Like, look what's happening in San Francisco and like no, this is uh

9:04 A whole different city and that's not. Yeah. Yeah. Okay. Okay. So someone posts this AI image. Uh, someone write to note, and this is actually five years ago in a different city. And this algorithm Helps. understand if this is a real if this is true, this note is true. And it's just people

9:20 Regular people Doing this. Yep, yep. I regular people uh who have signed up to be community notes contributors. So you know, there are a a few checks, like you do have to have a verified phone number, for instance. Um But yeah, at the end of the day these are these are regular people, uh, not necessarily professional fact checkers or anything like that. And yeah, that was like that was really important to us too. Like there was a question at the beginning to the point Jay was making of like

9:46 Well did did anyone think this was gonna work? Obviously it was kind of a crazy idea. We didn't know. If regular people were gonna be able to do this task. And certainly uh, you know. People had concerns about whether they would do it do it effectively. Initially some people inside the company were suggesting like, Hey, why don't you have journalists or

10:03 You know, some s some select group be the first participants, but very specifically we're like, no, that's like we're trying to move away from the idea of curated editorial decisions being made around this, this is supposed to be open to everyone. So there's very we very intentionally Try to allow All humans in.

10:20 There are people are randomly selected. Um, and that's important to it. Yeah, feeling fair, feeling open. Filling. Trustable. Yeah. And again, it's just like this sounds like the holy grail of understanding what is true and it actually

10:33 Works. And worked so well that Meta recently as you all know, uh decided to uh adopt this exact system for them in instead of having it tens of thousands of fact checkers reviewing things. One distinction that I would make, which may maybe can come off as nip picky, but I think is important is

10:51 Uh community notes adds additional context. It's not Fact checking necessarily. Right. So So there are cases where the post could be true, but maybe it's just misleading because there's there's no context. Uh or the missing context.

11:05 Uh and you know, we cover those cases and I think that's That's a kind of an important distinction. We also we just have the philosophy that users should be able to make up their own minds, right? Like here here's the here's extra context, take it or leave it, right? Yeah.

11:18 What I think about it, you shared this with me, this example of a Uh a picture with a with a with a cat. And somebody's community network was just that's a dog. Or is it the other way around? Yeah. Yeah, is yeah uh.

11:31 Palestinian boy shares his bread with a dog was the post, and it's a picture of th this cat, right? So like obviously this particular note is not super necessary'cause it just says that's a cat and links to Wikipedia for cat. Uh it's kind of a good example that uh That Like the system is this is not something a f a professional fact checker would ever write or think would need fact checking, but

11:55 Yeah. It's proof that the system is really run by the users at the end of the day, uh and and add some comic relief, I guess. Uh And you know, that's The note is correct. And I can you know it's important. When does a post

12:10 get triggered to even be considered for community note? Is there like a threshold or is it just you can write a community note on anything and people decide what they want on how's that work? So every post is eligible for notes. Um it's and that was again another really important principle. It's like It we shouldn't exempt Uh Elon, we shouldn't exempt government figures. We should like everyone, even advertisers can get note. So any post on the platform can get a note. And if you look in practice, Um, you'll see notes appearing on World Leaders, on Elon.

12:40 On ads. on media organizations and on obviously like just regular people using social media. But yeah, the idea is really that it's an even playing field. For a note to be proposed, the person proposing it has to have earned the ability to write notes. So there is

12:58 there is that that aspect where you have to like earn in to be able to do this. And the way you earn that ability. Is is through your ratings. by demonstrating the ability to help identify notes that are found helpful to a broad range of people. So basically like If you have an ability to to sort of see and know, recognize what's up with a lot of people. then yet you have the ability to start proposing notes.

13:20 I actually sign up to be uh What you what do you call these people? No no no taking. Contributors. No contributors. Yeah, yes. I've been rating and I haven't achieved. Nice. Can write notes yet. Uh yeah, it's not super easy. It takes it takes some effort. Are there stats you can share about the scale of community notes at this point, especially things that might surprise people? Yeah. Um I mean the service is growing rapidly. So there are hundreds of notes per day.

13:44 And to put that into context, I saw some stats recently from someone at UC Berkeley saying there there was something like ten fact checks, traditional fact checks a day. So In contrast, there's hundreds of notes a day. Um that are getting shown. They span a huge range of topics from obviously politics, news

14:02 Um out to entertainment, sports. Gaming, just whatever's going on that day. In addition to there being hundreds of the hundreds of these individual notes They can also be matched to multiple posts. So if someone writes a note on an image or a video, like let's say it's AI generated or

14:18 Something like that. that note will automatically be matched to all posts that contain the same image. So you can have a single node matching to thousands of posts. And Over the let's say the last the last year, twenty twenty four.

14:33 We had something like ninety five thousand notes that were seen about a thirty billion times. That's more than double the prior year. Prior year was something like thirty seven K notes in Fourteen billion times. So That rate is increasing dramatically. I mean, you think about like thirty billion views.

14:50 That's a lot of information that Is getting out there that might not have been out there. Otherwise, which is pretty cool. And The part of the reason it is expanding like that is the contributor base is expanding. Um there's something like nine hundred and fifty thousand contributors around the world.

15:07 That's Yeah, nearing a million People Making this happen. Um, which is amazing.

15:13 And I'm one of those, right? Like I count as a contributor. Yeah. Okay, okay. If you're signed up as a contributor, then there's more people on the wait list too. Uh so there there's plenty of had room for more growth. R regarding the the like the matching, uh the on on v media and URLs, um I think that's a huge way to get extra coverage. Also, I I do think uh we've been very careful to make sure that those matches are

15:38 Uh precise. Because I think one thing that people love about community notes compared to Other types of fact checking. is that actually the notes are custom written for the particular claim you're seeing, right? So so often

15:49 A fact check warning would just say something like You know. Get the facts here, and then there's a link to some generic page about voting, uh like uh information. Uh, which is you know, it's so so not helpful to have the the information behind a click. So

16:03 So pulling the context Top You know, so that you have zero clicks that you need to make and and keeping it specific is so important. I uh one feature I love that I imagine you guys thought deeply about is If I like the post in the past, I get notified later if a community note shows up so that

16:18 I'm not like remembering this false information. Yeah, I mean we we try to make notes as fast as they uh as we can. So we want them to appear instantly if possible, but Inevitably there's gonna be a time gap between when a post goes live and when people figure out what's going on and when they get the note out there. And so we send those notifications to try to close that gap. Um and yeah, we get we we get a lot of love for that. We see

16:42 Just people take screenshots and share them, they're excited about it. Um and it's also a pretty cool example of something you can do On the internet, in the social media world that was difficult in kind of like a print or standard news world. Um where you would see maybe a correction like the next day in a corner of a paper

16:59 But it was hard to read. Here you're getting a thing about it. If you've If you've engaged with the person and that shows up. Uh one user feedback. Uh point is I'd love the push to just tell me, here's what you got wrong. 'Cause I I find that I actually have to go into it and like read it. And I feel like the push could just be like here's

17:16 information. Here's more context to this thing you like. We'll go take a look at that. There we go. Live user feedback. Nice. Okay, I want to get into the origin story of this whole thing, but two more questions'cause we're on this thread. One is what what's the kind of the threshold for a note to show up on a note? Is that information you can share? Just how does that work?

17:35 So Just because of the details of the way the algorithm works, it uses this machine learning algorithm, you know, called matrix factorization where you know, we fit it with gradient descent and whatnot. The threshold is you know It's it's it's zero point four, uh on this, you know, mid up scale. I mean in practice what it means is Yeah, basically uh uh

17:58 a majority of people uh if there is a polarized divide relevant to the notes, you know, obviously some notes are not about politics or something polarizing. But if there is, then a majority a a sizable majority of people on both sides would would generally need to Uh find the note helpful. Uh uh uh And then

18:14 There are these uh there are other rules that come into play beyond that main one. So Yeah, even if it's above that threshold. Um, it might get filtered out if um there there's a separate algorithm that's looking at agreement between people's incorrect tags. So like maybe maybe people found the note helpful but incorrect. Right, like it happens, uh, and then those cases

18:35 It doesn't matter if it's above the helpfulness threshold. So is this point four is uh that's probably the wrong way to think about it, but is it forty percent of People that normally disagree agree. It means nothing like that. It's just like on some arbitrary scale. Okay. Okay. Yeah. Yeah, if we changed random other things about the algorithm, that number would also have to change to an equally seemingly arbitrary number. Yeah. That we We arrived at

18:58 add some like no some numbers like that by gauging user feedback. So we we could share a lot of notes with people, get feedback on which ones are helpful, and there's sort of just a line emerged about Yeah indicating where it's not. You know, where the where things go from like questionable to pretty clearly helpful. Yeah. And And it is set right now, by the way, to be really conservative, I think. Uh we We just

19:20 are pretty particular about quality and we really want no quality to be really high. I think Uh uh. I think Keith and I both believe that we live or die based on The cordie.

19:31 Uh of the notes at the end of the day. So So it'd rather not show a note that may be good, but we didn't have enough signal on, um, then the other way around. That makes so much sense. Like I've never seen a community note that is wrong. And breaking that

19:47 promise is a big deal. So I completely get why you guys are super conservative there. Uh okay, two more questions along the time,'cause I'm just curious. These weren't on my list of questions asked, but I feel like people wonder this. How many notes are written versus end up showing up? And triggering on it on it. We probably show about eight percent of notes that get proposed. Um

20:08 I think that's it's been between it's a seven percent and ten percent or eleven percent, something like that over time. The number can vary a little bit. Um And as Jay said, there are un undoubtedly and you can see it, there's there's clearly more good notes than we show. But The goal is to hold a really high bar. Like we want to show a note.

20:26 When it's gonna be helpful. When it's not gonna appear you know, bias and undermine trust in the system. Like we want these to be neutral, informative, helpful. And Um you know as Jay was saying, like we view the worst possible mistake as showing a bad note.

20:41 'Cause that's gonna undermine trust and the trust is is w is why people want the product. So So yeah, we the bar is there. And you know, like I said, there's there's clearly some um some in that remaining, let's call it ninety percent that are good. And then there's a lot that are just like not that great. And there's some that are bad. And If you write ones that

21:01 are bad, which bad being defined as people who normally disagree find the note not helpful. So it's like the inverse of the ones we show. If you write one that people you know, people normally disagree, find not helpful, you actually will ultimately lose your ability to write and have to earn it back. So That that range, the other ninety percent is a mix. Sometimes people look at the number, they're like, Oh, why don't you show more? It's like, Well, you probably actually don't really want us showing most of those. It's

21:26 The the The gold here is that The system is able to filter out the good ones. That makes sense. Okay.

21:34 There's many people that are very polarized, like very disagreeable with a lot of things. How did they filter into this algorithm? How do you deal with people that are like super antivacs, super Jan Six, like all these Ver extreme. Potential views.

21:47 If people really are so polarized, uh that there there isn't agreement uh among people who typically disagree. Yeah, it's possible that this is one of those notes that Might be correct, but But just Wouldn't be useful contact. It wouldn't be

22:03 you know, helpful to show uh as as context. Maybe maybe it's about acclaim. That people have you know, uh really entrenched opinions about and they've read hundreds of things about it already. Uh right, like probably probably this is just uh not gonna improve people's understanding. It's just not gonna be a helpful user experience.

22:22 So it might not be the worst thing, uh, in those cases to not show the note. Mm. People a few years ago were pretty pessimistic that maybe fact checking never Changes people's Uh

22:32 you know, understandings about what's true. Actually there have been external um studies, you know, run by people totally independent of us. Uh who have found that If you take a community note uh or post With or without a community note.

22:47 That actually people's Agreement with the core claims in the post does. Change. If they see it with the note versus without. So we are having an impact uh on this thing that people previously thought was maybe

22:57 Not so easy to do. Um and um so d it's nice to focus on the cases where there is the bridging agreement. I would also say There is this reputation component to the algorithm as well. So if you consistently rate notes in a way That

23:12 Is counter to the the bridging base consensus, then we'll stop counting your ratings, right? So Uh You know, if if you're the kind of person who constantly rates bad notes as helpful, uh uh we we do filter you out. So so there's a difference between those types of people versus just the The good but polarized ones.

23:30 Yeah, I think one philosophical thing that's important is that We want ever all of humanity to to participate. And sometimes people are surprised by that. They'll be like, Oh, aren't there people who are like, you know, shouldn't be doing this or like there's su you know, I don't they're Their thinking is so extreme or something, maybe they shouldn't participate, but

23:48 Our view is it's actually we wanna have all of humanity here because if we have all of humanity we can We then have the data to understand what notes will be helpful to actual humanity. Yeah, we can we can better model that, better better understand and better show those net. So It's advantageous to have people who have all sorts of points of views. And

24:10 Uh we don't expect that every note will be loved by every single person. Um, you know, that's kind of an impossible bar. But we do intend to show the notes that like eighty percent of people are gonna, you know, read and say, Wow, I'm gonna knew that. And so

24:25 You know, in that sense it doesn't matter how you know maybe Extreme someone views uh person's views as it's still great to have them in the program. Um so you know no matter what your views are, please sign up and participate. It it helps identify what's really helpful. Cool. And we'll link to people if they want to actually sign up so they know how to do this. Something we didn't actually touch uh sp specify. These are all volunteers, and I was getting paid.

24:47 To be doing his notes and voting, right. Yeah, it's totally based on intrinsic motivation and and we think that's a great reason to be doing it. Um when you talk to the most active contributors, a lot of them they just they wanna have better information out in the world and that's A great motivation. So yeah, that's why they And you know, if you If you think about like for these people, the impact they can have, it's kind of

25:09 Now Uh when we first launched US wide, this was like in tw twenty twenty two. A note appeared on a White House tweet. And the White House deleted the tweet and reissued an updated statement. And like

25:25 Like imagine being the person who wrote that. You probably have like twelve followers. Your your posts probably get, you know, a couple of likes. And here You just put a put a note on the White House and they change They're public talking points based on what you did. Like that is an incredible amount of impact. So

25:44 You know, it's Y you can see why people are motivated to do it when they care about what's going on in the world. It um You know, you don't have to be a big well known person to Shape. The discourse and information.

25:57 Well in a in a way that's helpful. It's insane. Like There's so much to love about this. One is just the meritocracy of this whole operation of just anybody can That is true and correct can participate and have impact.

26:10 Also just shows you how much information we get that is just wrong. Like we had no idea. How often we see things that are wrong. And now we do. Working on this product has made me realize just how many things I used to trust kind of by default that now I look at more skeptically. Yeah, skeptically.

26:26 Definitely meme these days. Okay. Uh, before we get to the origin story, is there anything else along these lines you guys think might be really important to share really Really interesting. Sure, I guess one other thing just um

26:39 Is is that Although we don't actually use the fact that a post was noted uh in the core ranking algorithm, um which you know, we we think is uh a nice property. Uh, there is a really big impact just organically, meaning not from the algorithm, but just from user behavior, where people will Like and reshare or you know, quote.

27:00 uh posts way less uh when when notes are applied. So Just I don't know for for people out there who typically run A B tests on big uh, you know, platforms you may already be familiar with this, but like one percent is typically an awesome effect size.

27:16 For any sort of algorithm change. We saw more like thirty to forty percent. Engagement rate. Drops uh for likes and reposts. And an A B test we were in.

27:26 Uh when uh comparing uh Showing a post to with or without a note. Which is just crazy big. Um And then and then if you actually look That's that's just an A B test on the engagement rate, so that's not the network effect. If you capture the

27:39 The overall network effect of how uh posts uh you know is spread less by That person's repost, uh this go feel like top line with a difference in differences approach. different multiple different external research groups have both found consistently that There's like a fifty or sixty percent drop in total

27:57 Reposts, which is just nuts, um, after another is applied. So It's having a really big impact on spread, actually, too. That's uh so ins Incredible impact. AI image of something false would just go crazy.

28:15 on Twitter and did before community notes came out and now What you're saying is just adding that context, not actually Like you're saying the algorithm doesn't demote it if there's something Incorrect. It's just people are like, Okay, this is false. Why would I want to retweet this? That makes sense. Correct. Right. Yeah, the the notes just totally take the wind out of these stories. So like the thing will be going viral, note appears.

28:36 Resharing drops fifty to sixty percent, and like that's it. Like it just you can if at fifty to sixty percent per generation this the virality quickly goes to zero. And and by the way, there's uh I I have very mixed feelings about this next one. Uh, but authors become eighty percent more likely to decrease or sorry, to delete their post after they get noted.

28:55 Which Okay, that's great'cause like less less misinfo out there. But I'm paying about because those are usually the best notes. Like if if the note was so just good that you had nothing you had no other option but to delete your post. Those notes don't get seen by other people, right?'Cause that is that's how you post.

29:14 There's an argument, by the way, that like Seeing that just because You might see the same misleading claim elsewhere off X or somewhere else on X, uh Yeah, it might be good to actually show up. To better to have seen the post with the note than not see it at all.

29:28 Yeah. Unsure about that claim. That is so interesting. Yeah. Uh yeah, I could I'd be so sad if I was that community r notewriter and just

29:37 Oh man, it's so good they just can't even keep the post up. Okay. So Coming back from Today's world where you're this like

29:47 Small amount of code. Is changing the way people understand the world. And what they believe in making the White House rescind their announcements. Zooming back to the beginning of how this whole Project started.

29:59 Uh what I heard just briefly is Uh Keith, you were just kind of Tired of managing PMs, you wanted to just work on something yourself. You wanted to work on something impactful. Away from corporate BS.

30:10 And you basically just started looking for something. That was Impactful, important, and you found this. Talk about just how it all came to be Beginnings of the story. Yeah, so I mean for me the beginnings actually go back to why I joined you was then Twitter. Um in twenty sixteen. I was uh at a startup

30:28 And we were we'd had some acquisition offers and One of them was from Company Twitter. And Uh, it was twenty sixteen, it was the middle of the election between Donald Trump and Hillary Clinton.

30:42 And There were like something like three televised debates, but every day there was a debate happening on Twitter and it was very clear like This is where people are talking about these things that matter where where I you know

30:55 information is being shared, where there where ideas are being formed. And As a user. It was obvious that I could get good information there.

31:04 But uh it was also obvious that there was kind of questionable information floating around. And I remember just looking as an outsider thinking like wow, like this is a really hard problem and it also seems really important. So we ended up going to Twitter. And the company was in a turnaround at that point. So like my first three years was just helping to get the company growing again. Um

31:25 Yeah, working on on everything that was the consumer product. Um, you know, getting user growth. going back and people wanting to work there again, et cetera. Um but a few years in I was

31:37 reflecting on what we had done. Um, you know, I think we had done a lot of good work getting momentum. Uh going. But It a and people in the us and in the industry had tried things to kinda deal with with misleading information. But

31:52 Nothing was really working. Like it was obvious nothing was working. Um Nothing could handle the scale of the problem, nothing could handle the speed. And a lot of people just didn't trust the existing approaches. The this the existing approaches were I there fact checkers or internal trusts and safety teams making decisions about what

32:10 was or was not misleading. And like A lot of people just didn't want or trust. that to be the way this was decided, which is very reasonable. And so Yeah, I'm I'm looking at that. I was I was still managing a large PM team.

32:24 Uh You know, that's a whole story in itself. I um I felt like I would that job required a lot of energy in and I and I didn't feel like I always saw the output that I wanted to see from it. Like I didn't see the change in the product I wanted to see.

32:41 And you know, I was contemplating should I go start a company? Uh should I do something else? And I kept coming back to this problem. I'm like, Man Like the w how is the world gonna deal with the with this information quality issue of like what we get. on social media, wherever we get it.

32:57 And I'm like You know, I'm at I'm at this company where you can make a difference on this problem. Like Why not go and try some crazy ideas and see if like one of them might work. And so

33:09 I came back, I had I had a kid, I came back from paternity leave, I went to my boss K Von, I was like, Hey K Von How about I just stop doing my job? And I go work on this instead. You know, this being try some crazy ideas to see if we can deal with mis with misleading info. He was stoked, uh, and so I went off and started working on that. Um, you know, it started with

33:30 Just reading W any research I could on the the problem in existing solutions, what was or was not working, or what were the issues. Um, and then into prototyping and then, you know, it ultimately led to us. building and piloting this idea that became community nuts. Amazing. Okay. I have so many questions and we're gonna keep going through the story, but

33:50 Uh when you joined Twitter, what was kinda the uh it was called Twitter at this point. I'm gonna try to call it call it X now, which I know is the Important to your boss. What uh era of Twitter was it at that point? Like it was Kavon run uh joined or And who is the CEO, because there's been many. Okay, yeah. I started I I came in December twenty sixteen. So Jack had rec relatively recently come back as CEO to turn the company around. And just to give you a sense of like the state of the company. Um, something like a third of employees were leaving every year. So just imagine that like a

34:19 A third of your team gone every year. Um, you know, the stock was in the toilet, um, the t product was not really growing. Um, and so Uh Jack was working on a turnaround, um, and K Mon was there already. K Mon was running Periscope and a bunch of uh video stuff. And You know that

34:38 D group continued to you know, Jack Jack was there um up through the start uh The the community notes and Bird Watch project. And uh yeah. Okay. And it was called Bird Watch. I I don't think we've used that term yet, but that's an important point. It was called Bird Watch initially.

34:55 Yeah, so it was originally called Bird Watch when um uh when we started the project, but obviously somewhat famously The name changed. Along the way. Yeah, maybe we'll just tell that story real quick. I know we're zooming into more, but just

35:08 I I have this uh Twitter thread that I saw with between Jack and Elon when they're Debating what to call it in Elon's like Birdwatch sounds creepy. I want to change it. Uh is there anything there you could share? Yeah, the story there the story there's kinda funny. Um You on

35:24 Elon came in, acquired the company, and you know, we had just launched the product relatively recently in the US. It had been in pilot for a year. But we had just made it available US wide. And he I guess he'd been seeing the notes and Um He this is soon after soon after the exhibition, he DM me and he was like, Hey

35:43 This community notes thing is awesome. And I was like, Oh, I'm glad you like it. Like let's, you know, talk. And so we talked the next day. And he kept referring to it as this community notes thing. And I was like you know. It's interesting that you keep calling that.

35:57 calling it that because that's actually the very first thing that I called it. Like the very first Figma markup I made. depict this thing was called community notes. It just I don't know why, it just felt really natural. And so that's what we had that's the first prototype we had tested. Uh you know, later the project changes changed its name to Birdwatch, but

36:18 You know, Elon was like, Hey. Let's just call it that. And so The next day we just changed the name. And

36:25 Uh you know, it's a it was uh it's always Uh you're notable for the team when you change your name, but really the the team was excited about it. I think it is a much more understandable name. Jack has made fun of it calling it uh like the ultimate Facebook name or something like that. But the most boring Facebook name everyone. Which is funny because they're now you know launching community nuts. Um But uh but I think it is a very understandable intuitive name and I think it has served the product really well. I there's there's a reason it was the name in the very first mock up. Yeah. I think descriptive names just make sense. Uh

37:00 This uh connection with Elon and uh I wanna talk later about just how you've dealt with so many strong personalities over and kept this alive throughout so many changes, but Before we get to that Um You. You did something that I think a lot of

37:13 Product leaders, engine leaders, just people that I've managed people dream of. Give up all this power in air quotes and career trajectory and influence. And just like forget all that. I'm gonna go back to just building something awesome.

37:28 Small team. Is there any advice there that you could share from that experience that you think might be helpful for other leaders to share or to hear? To help them maybe do that same Jump because that's really difficult in practice. Easy to talk about, hard to do. Yeah, I think it is a difficult jump. Um I've I've done it a bunch of times in my career and I've always been very happy with it, where you know I started with a small team, that it kind of grew into something bigger.

37:53 And then I was like, you know, this is like we're kinda dealing with a lot of big production stuff, teams really big. I wanna go back to doing something like crazy and new with a small team again. And so I've kind of done that like saw teeth leap. Um a bunch of times. But it can be hard because certainly the natural like the classic career path

38:11 Yeah. sort of I don't know, rewards or uh You know. running a large organization or being a manager or things like that. But I think at the end of the day, you gotta work on stuff you love, you gotta be having fun, and you I I think people wanna be having impact.

38:28 And I think there's one myth that that can get in people's way is the idea that the the more people you manage or something, or the larger your scope is, the more impact you have. I definitely do not think that is true. Um If you just look at I mean look at community notes for example. If I had stayed running

38:47 Oh, large consumer PM team. Like what would I have produced? Like Sixteen more pages of OKRs? Like I don't know, you know, a bunch of documents? And I think building community notes has had way bigger impact on the world. It's become the industry standard for how to deal with this now, which is super cool. People love it. It's the first thing that is plausibly

39:06 Dealing with the the internet scale, you know, issue of information quality. Um You know, I I think it's unquestionably a bigger impact than I would have had if I were just do whatever, doing s some standard management track thing like I was doing before. And

39:21 I think that's true of so many other uh, you know, small companies and and start ups. I I was just reading Some someone screenshot it. I think it's Blake Sholes uh LinkedIn the other day, he went from like director of coupons or something to building the first uh

39:36 Supersonic. Oh yeah. From group on. Yeah, yeah. Yeah. And I you know Those stories. are everywhere when you look. And so I definitely have found that, you know, for me, I love building hands on. I love trying crazy new ideas. I I love the zero to one experience. It's fun to scale things up too and it can be fun to operate at you know, at scale, but

39:55 Um Doesn't you Yeah, w our this team is a good example of one that operates at a very large scale, but that is still very small. Yeah, I think the way you guys operate is What

40:06 More and more companies are trying to do. Uh, remove middle management layers, create small teams that just execute and build impact and Just like I C's. Uh whenever I say I see, I'd have a comment on YouTube where like what is I see? So I'm just gonna explain individual contributor, non manager. Is when I say the word I see.

40:23 So let me follow this thread and When I asked people about how you Set up the team. to operate effectively and And protect it.

40:32 Initially. There's this term thermal that came up a lot. They it was like a thermal team, if that's how you describe it. Yeah. What is thermal? Yeah, so um Anyone who's works in a larger company probably knows that things can get kind of bureaucratic or bogged down, decision making can be slow, like there's these large planning cycles, people can like try to like take someone from one team, move them to another, like at random arbitrary times that can disrupt a project, like all sorts of things like that.

40:56 Um Uh, you know, our company, this is a number of years ago when we started this project, um, we had a lot of founders in the company, like Kavon is an example of a founder who was helping to run the company and He had this idea like Hey, why don't we create this program call it thermal? where we could have teams that were somewhat isolated.

41:15 From that. They could run on through their own process. They would have like one clear owner The team would be entirely dedicated to that project. And

41:26 uh we would just sort of like repeatedly make funding decisions as to whether to continue the effort. And so why was it called thermal, by the way? What was the idea there? I think the I think it was like an an old bird uh analogy of like thermals lifting, you know, the the birds on their wings. Twitter obvious through one point oh obviously had a lot of bird analogies, bless its heart. Uh and so you know, that was one of them. Um But the you know, the idea Uh I I loved the idea as someone who, you know, liked the startup environment. And so when we were starting this project, I was like, Hey, K on like.

42:00 Why don't we make this the first thermal project? And he was like, Yeah, let's do it. And so So we started with that way of operating and it gave us Yeah, from day one. Uh a lot of freedom and autonomy that I think was really important to make the product work. So just be very specific about what it w w what make it a what makes it a thermal project.

42:19 How do you set that up? And this is asking from perspective if a company wants to build their own something like this, what does that look like? Yeah, I think there's a bunch of key attributes. So key one one key attribute is Um, there's one clear driver of the project who's effectively like founder. Um, I guess I mean maybe you could have two or something, but like this like a really clear There's like driver of the project. And also

42:40 There's one clear decision maker that They go to. Oh outside of the team. Outside of the team. And that was true back when we started. And it is true now, like if we

42:50 I mean need something or have a question about something I talked to Elon. And It w you know it was like that from the beginning, it's like that now, and I think that's a big reason we're able to make decisions effectively, quickly, um, in a simple way. And it probably has to be someone very senior, not Yes, it needs to be someone. Someone senior who can make the decisions you need made.

43:08 Yeah, whatever they are. So th I think that's really important, that clear decision making structure. Another was one hundred percent focus. So everyone on the project is expected to be totally focused on it.

43:20 Um That Yeah. A lot of companies it can be easy to have people's attention sort of spread across a bunch of things. And it makes it hard to

43:31 Get stuff done. You'll go to like a you'll try talk to who you know, whoever that person is, you'll ask them for help on something and they'll be like, Yeah. I'll help you. I gotta finish this thing. You know, and it'll take me like a week or two and then I'll get to it.

43:43 Like a week or two delay totally changes the momentum of a project. When you know we were a hundred percent focused. Yeah, we talk we talk in the morning. It's like, Hey Jay, why don't we like try this thing on the algorithm? He's like, Yeah. Then like then you know, that afternoon or the next day we're looking at results. And so y because of that total focus, the rate of iteration Goes way up.

44:03 And then you know, beyond that there was Uh also just the ability to use whatever our own sort of like decision making process was. We didn't need to Right, OKRs W or you know, follow others like standard practices. Obviously like we had to make sure we were responsible. responsibly building the product and everything. But

44:22 Yeah, we didn't use need to use standard um the standard practices. And I think that's a another great example. Like OTR is I understand why They can be helpful. Um but They can also be

44:33 uh, you know, n not necessarily the right cadence at which to set goals. Like I don't I think it's really unclear that quarterly or annual goals are actually like the right pace. Like We would set our goals for what like we would set the goal for the next milestone that mattered and we would work on that. And when we reach that milestone, we would have an idea of what was coming after and then we after when we hit that, we'd set the next milestone. Whether that was two weeks.

44:56 A month? Three months? Like what whatever it was, like we set our own pace and and goals at that pace. And that just I think is a lot more natural for the development of something. The the whole OKR determination and planning process took longer than it would take us to pick a goal and then execute on it and finish it. How big was the team early on? Uh That you said it.

45:18 How many engineers? It started with just me and then we we when we Decided to build the thing. figured we needed about five. Um and we wanted it to be as small as we possibly could. It was clear we needed someone on M L doing scoring. Yeah.

45:33 It was clear we needed someone to do some client engineering work, someone to do back engineering work. Um there may have been like w you know, one or two other oh, we needed a designer and um a researcher to help us understand. The customer base. And they make sure we were we were building the thing. in a way that was actually gonna w to resonate with people. And so I think that was I think it was like back end, front end.

45:56 Um, M L design research. That was that was the original team from what I remember. Amazing. So one basically one of each function. A question I have for Jay actually is There's all this talk of small teams and Uh moving fast, but you know, sometimes you just need more engineers to build the thing.

46:12 Is there anything you've learned about just how to keep a team small while moving as fast? as you are and not need we need to hire more engineers. We need to hire more engineers. I think in the r in the beginning when we were iterating uh you know, what what should even the requirements be, um it was it was Definitely good to just have it. Like one animal engineer. But I think at at some point we got clear on what the goals

46:37 of the the algorithm should really be and we try. You know, we were real I think at the the very beginning it wasn't clear that we needed to build this bridging based uh Algorithm, right? The actual first algorithm. that I put into production was very focused on anti manipulation. It was this kind of page rank variant.

46:54 Uh but it didn't solve the problem of you know bias, basically. So if if there's some if there are more users on one side, the a page rank type graph algorithm can actually amplify those biases. So I think you know, after building that prototype and and getting data from that, it was clear That

47:12 You know, uh the the burning based algorithm was gonna be the way that we needed to solve it. And at that point Basically uh I set up a a bake off. Basically made you know, this this kind of like a the uh you know a cagle competition or something.

47:27 So that was like the key time. Uh where It was really important to pull in other engineers. That is such a cool story.

47:35 Uh I wanna follow that thread before we do that. You guys yell thermal. What does that mean? Is that like YOLO, like a version of the Okay, we're just gonna ship'cause there were thermal projects.

47:46 Ship it. Okay. Marketers, I know that you love TLDRs, so let me get right to the point. Wix Studio gives you everything you need to cater to any client at any scale, all in one place. Here's how your workflow could look. Scale content with dynamic pages and reusable assets effortlessly. Fast track projects with built-in marketing integrations like Meta, CAPI, Zapier, Google Ads, and more, A B test landing pages in days, not weeks, with intuitive design tools. Connect the tracking and analytics tools like Google Analytics and SEMrush, and capture key business events without the hassle of manual setup.

48:23 Manage all your clients' social media and communications from a unified dashboard, then create, schedule, and post content across all their channels. If you're working on content rich sites, Wix Studio's no code CMS lets you build and manage without touching the design. And when you're ready for more, Wix Studio grows with you. Add your own code, create custom integrations with Wix made APIs, or leverage robust native business solutions. Drive real client growth with Wix Studio. Go to WixSudio.com. Okay, so coming back to this algorithm, this is actually really interesting because I I've never heard any of this.

48:55 So I was gonna ask just what inspired this actual algorithm. And he basically did an internal competition amongst ML engineers. To see who had the most successful algorithm, Netflix. Pro uh contest style, cagle style.

49:09 Yeah. Yeah, I I think So I mean this particular idea of of Finding you know, content that is liked by people on opposite sides of the of a polarized divider who who typically disagree. No, this was not an

49:21 An idea out of thin air, right? Like I think Keith had found some of Chris Bill's work he had done. He had you know, made this list of accounts that were often liked by people who Uh you know, we're on both sides politically. Um there is um

49:36 you know, other other projects like uh polis out there that That look for agreement among uh Yeah. people who typically disagree, but I think the

49:46 Yeah, it it wasn't Obvious that Our project definitely needed to use that from the very beginning. But then in w you know, when you When you implement it and compare it against these other type like page rank seems obviously you know, it's designed to be kind of manipulation resistant. It's naturally

50:02 Naturally, like if you just have a voting ring of people who all vote them themselves up, then PageRing can filter that out very well. But Like that just wasn't the main Attack. vector, I guess. So we we we got it we had to get some real data from the pilot to realize that okay, the real thing Going on here.

50:19 is people are polarized. And uh and And so It was only once we got that uh the the real data from the pilot, that I think it was it was clear that the the bridging base algorithm was was um the direction we really needed to go.

50:34 I wanna come back to the way you operate the team. I hear that you run the whole team off a single Google Doc, that's like a four year old doc they just keep adding. uh goals too and bullet points. Is that true? There there is a very long running doc. Um, that has had to be uh chopped and purged because it was breaking Google Docs and Chrome at various points in time.

50:56 It's sort of like a note taking doc um It's really where we coordinate what we're doing. Um, the team meets on a daily basis. Um, we spend whatever amount of time we need to get on the same page about what we're building. It can be Yeah, we might talk about anything from You know, what's most important right now?

51:13 to you What are what should we work on next to What are we trying to launch right now and why is it not launched? Like what's in the way of launching it? And we might review uh new modeling or scoring algorithm update and you know trying to understand what's working in it, what's not.

51:30 Um, so we'll just cover whatever we want and um Or whatever feels most important. And we like you know, as you said, we set our goals very dynamically. Um, so it's whatever seems like the most important thing for us to work on now and next is what we spend our time on. Uh I think that served the project really well. Um, versus ha feeling attached to like

51:51 some kind of quarterly goals or something like we'll look at like What is gonna help people the most, or like what's the biggest problem right now? What are either one of those, and we will go tackle it. And we can We might change our roadmap. Yeah.

52:04 multiple times in two weeks based on what we see. So I'm hearing no Jira, no Sana, no Monday dot com. No. Okay. Yeah. I mean we we have to use Jira to like coordinate with some other teams. Like sometimes when we file a request, we have to make a Jira ticket, but no, I I am not a fan of heavyweight task management. I love like being on the same page, being able to keep most things in my head and having a really light way to write down the things that You know, I can't

52:31 Or the team can't keep in its head. Oui. We did use Asana briefly, um but My memory of it is that it spent You you spent more time in the meeting grooming a backlog of irrelevant stuff.

52:43 Then actually you know, talking about the the proper priorities. Um so I think it's nice in the Google Doc that if something becomes Relevant. Fall off without needing Explicit backlog grooving. Mm-hmm

52:58 So just to maybe summarise a little bit of How you guys operate that might inspire other companies to set teams up like this. Uh so I'm gonna go through a few things I uh you shared. One is one person in charge of the team, like the founder almost. They were like basically the founder of the team. Yeah.

53:14 They have one very senior essentially sponsor slash decision maker. That they interface with, in your case, Elon, no big deal. In other cases it could be the CTO, CPO. Someone like that.

53:25 Uh the team is focused one hundred percent on this product. and and and goal You uh keep the team very small. So you start with one person of each function when from an engineer back in. ML person, designer researcher. Yeah.

53:39 And then Uh Google Docs almost basically for your Project management. Yeah, it's basically running with Google Docs stop.

53:48 Don't use big complicated products. I think that's a pretty good recipe. On the Google Docs, you know, take it the people can do what they want. If they want to go for it, I think those those first ingredients are really are our key structurally. And then, you know, beyond that, it's a matter of having an ambitious goal that gets people fired up to Go do great work. Yeah.

54:07 Uh, awesome. I think there's a lot there that a lot of people kind of like think they should do when they set these teams up, but they don't actually do. And it feels like each of these is Just a really key ingredient to it to actually succeeding. It definitely really helped us succeed. I don't know that the project would be here if it was not for some of those a elements. That's that's a powerful statement. Like

54:26 This thing that has changed the way the world understands what is true would not have existed if you didn't set it up in this Specifically. Yeah, I think uh you know, I I don't know if I would have started would have begun the project had I not known we had sort of that structure, that ability to make decisions, the autonomy, the the speed, the ability to go fast.

54:47 And Yeah, working We s we started with that in one point oh and and it's been continued and if anything furthered in in X. I mean X as a whole company operates with a lot of those attributes. And

55:01 Um I think it's It's one of the reasons the product is successful. I think it's a big th those are big reasons why at least I Jay can speak for himself. I have so much fun working on this. Like I I I love Working on it.

55:15 You know, it's great to wake up every day and solve these problems, we get to you know, we get to do them efficiently. Make decisions quickly, build stuff that Helps a lot of people. It's it's awesome. Yeah, this this like uh whether thermal or elon way of operating is definitely more fun and

55:30 Like that combined with the the Awesome mission. is super important for internal recruiting. Like I remember uh Like when I was first chatting to Keith about this back in early twenty twenty.

55:43 Yeah, I had another project. I was you know, uh work on a few, but one one was like personalize the number of push notifications that we send. Uh and it was It drove a lot of DAU. Um without like losing opt outs uh uh significantly.

55:58 Uh so you know, that that was like Setting me on track. Uh or you know, if I had kept working on that, I could have probably gotten a promotion from that, um, with low risk. Or I could take this huge career. I mean it's not as big a a career risk as like joining or founding an actual external startup, but

56:15 There is no career risk, I guess, and joining a team like this. So so just I think all of the same aspects of recruiting that apply to external startups and apply internally and and you know, if you can have Um an exciting vision that

56:28 Uh is key. Related to that and your list, Lenny. One thing we missed that's super important. Is that On this project and I think of

56:38 Successful Projects like in startups. Is that people are self selecting to join. We did not assign anyone to this project. Like People

56:49 reached out to join or they applied to join the job. You know, I and the team interviewed every single person that joined the team and were like, We want that person on the team. They want to be on the team. And so people are totally bought in to The goal

57:04 mission, the way the team works, the other people they're gonna be working with. And that makes a huge difference. So I've like A great time to do that is at the start of one of these things. Like don't if you're gonna try something crazy, like I would uh it's gonna be tough if you're just assigning random people to it, but if you let people opt in. And self select, you're much more likely to be successful. And one thing that I have observed at X, which really surprised me was

57:28 Was that this is also possible at a large scale. Yeah, one of the things Elon did when he bought the company was he basically asked people to self select to stay. Like you had You had to click the button. And And he sent an email out that was like

57:42 Hey, Twitter two point oh. Fork fork in the road, right? Fork in the road. Exactly. It's like Twitter two point oh You know, now X. It's gonna be hardcore.

57:52 We're gonna do ambitious things, you're gonna work your butt off, you know, what and You had to click on the form and say, Yes, I want to join. And I think that was really important. For the company

58:04 Because You want people to opt into that. You want the people to be saying, like, Yeah, that's what I wanna do. And the company's gonna be a lot more successful. If people aren't sure, it's like better for them probably to go do something else and where they're naturally more aligned and happier and I thought that was a great approach to taking a large company and getting it down to people who are really excited about you know, working together on a on a mission. So, you know, for us, we did it from day one, which I think is an easy way to do it, but it's possible to do it later as well. I love that you described it as fun.

58:35 And I think a lot of people when they see Elon Laying off a bunch of people, being like very hardcore himself. It's People don't imagine it as a fun place to work. And it's clear how much you guys love working on this, like how fun it is and how interesting it is.

58:50 And it's interesting to hear that,'cause I think a lot of people don't feel that. Externally. Is there anything else along the lines of just working for Elon within an org Elon runs? that may surprise people about just the way of working that's interesting or surprising or Or you think other companies might want to think about adopting?

59:07 I've always liked lean teams. But This has made me my my experience at X has made me change the way I would I would think about running a future or you know if I were to start a company and change the way I think about starting that company, it would be even leaner than I would have made it before. I've been amazed with just how much the team

59:27 is able to accomplish Um with a small group and I think because of a small group. Like When shortly after the acquisition, um yeah, we had this product called Spaces. Um it was It had been in the product before, but it was it was

59:41 Pretty small scale. And Elon wanted to run these large spaces. I forget who the first people he was gonna bring on were, but he was gonna be there, you know, ultimately these things have gone on to host politicians and things like that. And he's like, Guys, we gotta scale this up. I forget the numbers. I we did need to be able to scale like a million people or something like that. I'm getting the numbers wrong, but you'd be able to scale way up.

1:00:03 This is the kind of thing at one point oh that would have taken a year if it had ever happened. And the team did it in like two or three weeks. And Uh It was really exciting and inspiring to see. Like we I didn't work on that, but I watched it from the outside. I'm like, wow, with this tiny team.

1:00:21 motivated behind a big goal that was like, hey guys, it's not like are we going to do this, it's you we are going to do this. They got it done in two or three weeks. That must have felt amazing for them. It was certainly exciting to see. Um but it's uh I I've definitely come to appreciate just how

1:00:36 Um lean that something can be and and not just get by, but actually thrive because it's that lean. I think the point you made about people opting into that is important'cause I think a lot of people hearing that be like, I would never want to be Asked to build something like that into weeks.

1:00:51 And I think a lot of people do and w we'll we love that kind of experience, especially working with Elon, especially shipping something at that scale. But I think there's an important element there of just like, Okay, I I don't wanna do that. I have other things to do in my life. Other than chip spaces. So I think that's a I think that's a key point you've raised of just there's an opt in step. Totally. I think the opt in is important. And it may even be that you want to opt in one part or you know, at one point in your life and maybe at another point in your life.

1:01:16 Something else is better, I think, you know, whatever it is You're choosing to do, it's nice. to to be opting in to feel like it's aligned with how you want to spend your time. Something on my mind, and I don't know if you guys want to go here, but it's something I think a lot of people think about is when Elon came in, he let go of eighty percent of folks, and everyone's just like Twitter is dead.

1:01:35 It's all gonna fall apart. There's no way they can run this thing. with that small of a staff and clearly it they were wrong. Clearly it's working great. It's like Becoming a like a massive deal in the world and continues to grow.

1:01:48 Is there anything about that that you Or surprised by or anything about just like how it's It continues to operate so well in spite of that big shift. I think the The the leaner team, the

1:02:01 the reduced kind of like process in bureaucracy is the big reason it does Move Yeah, as fast as it does. Um It's easier to get stuff done faster here.

1:02:12 And Uh Yeah, I mean I think that's uh I think I think it's that that shrink shrinking is actually a big reason for the Increased pace of launches. The increased pace of experimentation.

1:02:27 one thing that I noticed that uh as a result of that is the people who are here they seem to all really feel like owners. Like they take like they take the sense of responsibility that an owner takes in the product. Um they'll try to track down what's Wrong. Fix whatever is needed. Um, jump into it

1:02:45 any to to help build or fix improv any system that needs help, even if it's outside of their space. And there's the flip side of that too, for people who've worked at big companies, they may have experienced this thing where Where there's like a nut you wanna change some something in some other system or product. And so you reach out to that team and like maybe they're a little resistant. They're maybe like, Oh, we'll get to that next quarter. They have their own goals to hit. Yeah. Yeah, exactly. Like they don't really necessarily want to help you or they're busy.

1:03:12 Here, you're like Hey guys, we need to do this thing with that other system you work on. And they're like, Great. Here's the code. Here are the docs. You know, s send us the fab if you have any questions and we'll get it in. And it's just the thing y you can just jump in and and get it done. And

1:03:28 That kind of collaborative effort, like the sense of like shared ownership, I think And from my experience came from Or result it was a result of the shrinking of the team down to people who You know, wanted to be there and work together to build this thing. So I think that's been

1:03:45 a really positive impact. It's not always easy. Certainly like a lot of people have a lot of responsibilities, but You know, they're here Because they're up for it. Yeah, I think one other thing that's key, uh is Uh when when you are forced to have such a small team

1:04:00 Yeah, deleting. Well, this is important anyways, but deleting code is more important than writing it a lot of the time. Uh So I I think So often, maybe due to promotion incentives or just regular human tendency. You uh

1:04:13 You know, engineers have a uh a tendency to add these little incremental wins that actually ha Add you know, more of a long term maintenance cost than is clear because you just run a little one month A B test, you see this. you know, a significant win. And you don't realize the maintenance burden you just

1:04:29 Added uh to your team for the rest of eternity until you turn the thing off. So I think there's A lot to be gained and and you get forced to do this, um uh, by the way, when you have such a small team is just Deleting, you know, uh auditing Auditing parts of your system and deleting the things where the maintenance costs is

1:04:47 Um worse than One than the games. So I think th th we did have to do this across the company, um after the the big layoffs and Um, you know, systems are leaner now and and they can be worked on by a fewer numbers of people. That's an amazing point. I remember Elon being like here, we have to throw away the whole thing. We have to re-architect everything. It's stupid the way it's built. And sounds like that actually worked. Yeah. Well we didn't uh you don't have to rewrite everything from scratch. I mean some things we did I guess rewrite, but uh I I mean just even deleting the unnecessary cruft and keeping the

1:05:20 rest of the core system. Um That's awesome. I love the word creating kind of a formula to run these sorts of companies and teams. There's so much here. I wanna go back to the the building of the original product. We ca it kind of took us on a long tangent and an amazing tangent.

1:05:34 But Uh I heard a story of when you launched Birdwatch at that point. You specifically wanted to keep expectations very low. And there was like a gif in the thing and it just looked like clearly this is not uh

1:05:46 Ready for prime time. Talk about just how you did that, how you launched it in a way where people weren't like Um it's never gonna work. discipline, I guess you could say, about having the product prove itself at ever at every given s point. Um, you know, when when we built the first mock-ups, we had just these were just like pictures of

1:06:07 D you're depicting what community notes might look like. We showed those to people across the political spectrum. We saw Like, hey, people really like these, whether they're on the right or left. Like they seem very open to reading these community notes, even when they are critical to people of their own side. So we're like, all right, that gives us confidence that if we can build this, like if we can actually make this as a reality, it's gonna work. Then there's the question of like can we make it a reality? Like are will people in the real world be able to write notes

1:06:34 That are of this quality. And so Um yeah, we built we had an internal pilot Test version of this where you could like write notes. And

1:06:42 We first basically ran this through like an Amazon M Turk type of uh participant test just to see like if you just like Put some normal people in there. Like will they be able to write these notes. And it not they weren't all all those notes weren't good, but like it was clear that there were people out there who could write good notes. So then like okay.

1:07:02 This is possible. Like What will happen if we actually do this out in the real world? And like let's run let's run a pilot and find out. And so We took that pilot. The you know we'd run the M turk kind of test on.

1:07:14 And we wr released it to at first a thousand people. Tot you know, totally out in public. And We didn't know what was gonna show up. Like you can imagine The nose could have been terrible.

1:07:27 And Uh, and so we were talking like well what do we do? Like we're gonna put this out there, everyone's gonna have all these questions, they're probably gonna be really skeptical. Like and we know It might be a total dumpster fire. And so like what do we do to like set expectations appropriate? Like we felt like we could probably get there in the end, but we just didn't know it was gonna happen at first. We wanted to set expectations and so Like wow.

1:07:47 Why don't we just stick there's like the page where you see a post in the notes below, where like why don't we just stick a dumpster fire gift like on that page? And uh, you know, you go there You're like, hey, I you know, anything you see below here might just be a total dumpster fire. Um, at least it would show we we were aware of that as a possible risk. Um we in the end we did not do that. It cracked me up. Um, but we thought it was kind of like Oh, you didn't actually launch okay, this was uh just mock ups of it and every time I looked at the mock up I laughed. But

1:08:21 Uh ultimate we had so much to explain on that page, like what is this thing and how does it work? Ultimately like, okay, this is probably gonna like distract from the point. Um So we pulled it. So I somewhat I kinda wish maybe it had it seen the light of day at one point. But yeah, ultimately we kept it simple and we focused that page on explaining what was going on here. But again, you know, we um As has happened many times with Titans with the project. Um

1:08:44 Yeah, we put the pilot out there. And the notes were good. Like they weren't all good. There's it was a mixed bag, but like there was gold in there. And from the very early days with just a thousand contributors, it was obvious. That that people could write notes.

1:08:58 that were informative, that were neutral. That spoke to controversial challenging topics. And that if we could Just identify those from the rest. This was gonna work. Like it was gonna work as well as the very first mock ups we had made.

1:09:12 So that became the the the focus then is how do we sift out the gold from the rest. I remember uh there's a I think you may have shared with this with me when someone noticed you guys were testing this. And they took screenshots and tweeted it and I think Elon replied, like, This is cool. Yeah. Yeah. So in the in the very in the very early days when it was just a Figma prototype.

1:09:32 We were running these like user testing.com unerated studies. I guess one of the participants sent one to an NBC reporter who like wrote a bunch of stories on it. Anyway, like that day There's a lot of chatter about it on on the service. And Elon, this is like this is put this is back in

1:09:50 you know, time perspective. This is I think twenty twenty. So Two years before Any acquisition stuff happened. Elon is just a Twitter user. Building rockets and electric cars and other cool stuff. And

1:10:04 Stumbles on this thing. That d depicts the prototype that we've been testing, and he writes back Yeah, definitely worth trying IMO. And I remember thinking that was cool back then. And it's interesting to see like it he

1:10:18 He's obviously had a very consistent point on it. I think the you know the idea was appealing and he uh you know's obviously been a big fan of it, um in the product and been a big supporter of proponent. So Uh yeah, it was kinda it was kinda cool that it came from That support has been from the very early days before he was ever involved in the company. I love that moment. It must have felt really wild for Elon to be commenting on this Figma prototype retesting. It's cool.

1:10:43 It was cool. Oh man. So when we were preparing for this interview, I asked uh you guys what's what's the main thing you wanna make sure people get? and understand about why community notes has been so effective. And Keith, you specifically said that it was

1:10:59 The principles behind how you wanted to approach this and how you continue to stick to this throughout. And we'll talk about how you kept it alive throughout all these different CO changes of leaders. But just talk about

1:11:11 these principles, like what the actual principles are and why that was so key to it working out. There are a number of principles that I think when we first shared them with people at the company seemed maybe a little bit crazy. Um, but I think they are they are the reason the product works. And I think

1:11:29 They've been very important and we do. We come back to them regularly. Today. All the time. Probably the craziest one. Is just that

1:11:37 This thing. is going to be the voice of the people. It's going to represent the voice of people. It's not gonna represent The company's voice. So it is not a tech company deciding what shows, it is the people Deciding what shows.

1:11:51 And that had a lot of implications on the design. Like First of all. There is no we don't have a button that will change the status of the note. Uh of a note. So if a note is showing

1:12:03 Because the people have rated it. And found it helpful, it is gonna show. Like we can't change that. And that is the kind of thing that like at for we first proposed this, that's unsettling to people. They're like, Wait, so like something can go up and and like we, you know, the company can't take it down or uh you know w can't change its status, get it to stop showing. And we're like, Yeah.

1:12:22 And like it has to work that well. If it doesn't work well enough to do that, then it doesn't work. If there's a problem with the note, this is like one of our one of our key principles was if there's a problem with a note. That's so bad you want to do something about it. It's a problem with the system. Like we need to redesign the system. to n to be showing good notes. And so So yeah, we had to, you know, get everyone comfortable with the idea that there was no button to

1:12:45 To change the status of a node. Yeah, similarly, we as we talked about earlier, we wanted this to represent all of humanity. And so we didn't want to be arbiters of who Can come in and be a contributor. And he can't. Like so we if we we open it to Everyone, you just have to meet really basic objective criteria. Like you have to have a verified phone to help reduce the likelihood of having like

1:13:06 bots or things like that participating. But beyond that It's random selection. And it still is that way today. And you know, again that people will it took some time to get people comfortable with it. But I think the the f the fact that this

1:13:20 Is the voice of the people and reflects their Output transcript: through an open and transparent process is so key to Both why it is

1:13:30 Good. Like why it works? But also why it's trusted. So I mean that's number one. And and you it's you know, will I think will forever be at the heart of the product. Um Another one um that people thought was kinda crazy was transparency.

1:13:46 Yeah, we're like we're this d we we the previous approaches to dealing with misleading info. They it felt to a lot of people like sort of black box tech companies or media companies or leads to whatever making decisions. Like They people n people need to get comfortable with this. They need to trust this. So the whole thing has to be out in the open.

1:14:06 Like The code that decides what Note show has to be out in the open. All the data and ratings that that make it happen have to be out in the open. People should be able to take the code and data and replicate the whole service and

1:14:20 That that we have done exactly what we've said we've done. And they should be able to audit it. They should be able to go and look and say, like, hey I think this part could be better or like if they think we're biased they should be able to work, you know Work with the data and point it out and if they if people f have good observations, that should factor back into the code.

1:14:37 Um and This is again something that's kind of difficult to get people comfortable with. Um That everything is out there. You you can't cover anything up. But I think that's so essential to people trusting it. So Um

1:14:49 Yeah, I mean we we We set these out on day one. We go back to them constantly because we're we're We're always evolving the the product and we always like Gotta make sure every change is open. Like whenever we update the code, there's a or update the scoring system, there's an update in GitHub. The data is published daily, so you can download it and so

1:15:08 Um yeah, I think those have been those have been really essential to the thing working. Yeah. And and by the way, these these do not come without a cost, right? Like the the It's actually really hard from an end perspective to actually open source the actual algorithm that's running on the actual data. Uh Yeah, because the with the way large scale services like this are usually architected does not

1:15:29 you know, naturally lent itself to like being run uh like as a script by someone who's downloaded a TSV. So we actually have to take weird architectural decisions uh to make this possible in a way that probably wouldn't have been if we didn't start with this. assumption from scratch you would have had to maybe rewrite the system to make to To make it like this. What's an example that you're going to do?

1:15:50 For instance, there's a matrix factorization that we that we train. Right. Usually You had um Train a matrix factor is a you know, train your ML model once and then Um service uh uh

1:16:03 uh I guess with a separate service uh But uh you know, we didn't want to have people externally spinning up services, uh, you know, to to be able to repro replicate the um you know system that we had. So I I mean basically I don't think it would have been actually very cool if we had

1:16:20 Uh open source the code in a way that wasn't Actually runnable, I guess. Um you know, uh by someone Just yeah. At this point you can download Python code and and run a script. You do need a lot of RAM right now, but how can do it on one machine. How much RAM we talking about? Oh like five hundred gigs. It'll take like a day if you don't uh do anything special to speed it up. But

1:16:45 Cool. Possible is the key thing. And people have done it. Um Like uh Vitalik Peter and had a blog post where he, you know, talks about uh his explorations You know, making sure the algorithm really does what it says it does and No. I think just the fact that

1:17:00 A handful of people have done this, uh You know. There's there's uh enough people who have done it that there's someone Do you probably trust whose It was verified it. Yeah. Yeah. And now it's rolling out the meta. No big deal.

1:17:13 Uh I love just like as you describe these principles, just I can imagine a PM at a company being like, Okay, guys, here I wanna do this. Project. can be completely like there's so much idealism to it that rarely works in real life. Gonna be open source. You're gonna give it to everyone. We don't have actual control over what it's gonna do. Don't worry about it.

1:17:31 Gonna just change the way People see this thing that we've been very careful about. Uh and then it works. And I think that's very rare and it's really impressive. And what I'm hearing partly is that the sticking to those principles was actually

1:17:45 really fundamental to it working and not kind of bending over backward bending over when someone's like, No, no, no, we can't do this. What if we change those part? I think if we had if we had broken with any of those principles, like if there was anything black box, if there was you know, w whatever I I the product would be a lot harder to trust. And so I think it's because we've just stuck to them so cleanly. Simply.

1:18:08 that you know people people can't trust it. You've talked about a few moments when it was like wow the the White House changed their announcement because of a community note. We talked about the the dog is a cat. Are there any other moments that After you launched of like holy shit, this is working. This is gonna actually work. All along, you know, we saw it we saw it working. Like we were pretty we

1:18:30 We wanted to be confident whenever we ex expanded it to new uh you know audiences or new countries or whatever. Like we wanted to be confident it was gonna work. So we You know, maybe You'd hold her breath a little bit just to to see that it would do what we expected. But we always expected that. But that said there were definitely

1:18:48 Stress. Stress cases. I mean the one that comes to mind is the start of the Israel Hamas conflict s uh in twenty twenty three. October. Um That was probably

1:19:00 the largest deluge of misleading information I've ever seen shared on the internet at one time. It was like overwhelming. They're just the um number of photos and videos and whatever coming out related to that was it was insane. And Um, just to you know, to give you an example in the first I think it was like first

1:19:21 three days or something of that conflict. We had five hundred notes. The covering All sorts of different uh you know like out of context imagery, like people someone say like hey this is happening here.

1:19:35 Um, it's actually from like twenty thirteen in Syria. There were people making fake like battle footage in the video game simulator Arma three. So they're like not explaining the stuff looked really looked realistic. And You d unless you saw the note, you like wouldn't really know.

1:19:50 Um there are all sorts of claims about what was going on in the ground. And Yeah, that was definitely The the product was still pretty new at that point. We had only we'd expanded in the US um Less than a year before that, we had been rolling out

1:20:04 throughout the world that year, and then this large event happened and um Yeah, I felt like we were We were just enough prepared at the right time for the system to be able to handle that. Like probably One of the most important things we did right before that was launch the ability to write notes on images and media and have uh images and videos and have those matched to other posts.

1:20:25 Like I remember at that time thinking like Wow, I'm glad we launched that feature a few months ago. Versus still had it on the shelf. Um because it was really important in that conflict.

1:20:38 I think even It was like just a few weeks before we had launched a major speed up. in notes. Like when we when we first built the product, the number one focus was always quality. Like we knew We knew that that notes the product would live and die by the quality of the notes. And like that was the thing we could never give up about. We also knew it needed to deliver speed and scale, but we're like We will

1:20:59 Was get the quality in the right place and we can speed we can speed it up and scale it out over time. And we had actually just launched a speed up that took like three hours off the time it Um a d time and note needed to go live. And I was pr you know, I think a matter of weeks before that conflict happened. So again, like super glad that was out there. The

1:21:18 In the first few days of the conflict, the median time from a post going live to no showing up was five hours, which is like crazy fast. Typical fact checking is like two to four it's really common to s see it take two to four days. These notes were showing up in five hours and we're like We are so glad. We got those things out before this happened, um, it made the service a lot more helpful.

1:21:40 One other thing that was um I I think uh uh nice to Yeah, see working then was I think one criticism of community notes some people bring up is well, if you always need agreement from people who typically disagree, then in these super polarized settings, like that conflicts being like probably number one Uh then

1:22:01 Then you know, uh you wouldn't see any notes. But actually the reality was there were the you know, tons of notes about that conflict. Um so I think there is this kinda Um You know, nice

1:22:12 uh nice property where actually and maybe this is a surprising fact that there's more agreement out there across polarized divides than maybe conventional wisdom says, right? And peop and the places where people agreed were Really objectively true and verifiable. Like uh I guess maybe this is more true the more polarized the setting is, but Uh w where the ag the agreement actually

1:22:34 uh lands you an Uh basically notes that are very neutrally written, very focused on the facts and and easy to verify. Um information. I love just like there's this talk this talk for a while of just like no more f there's no more facts. Like nobody

1:22:50 believes there is a f single true fact anymore. Like er everything is subjective. And I think immediate notes is the exact uh proves the opposite. Yeah. Facts matter. There are facts that we can all agree with, even in the most controversial. Topics. Yeah. Yeah, we we saw this really from day one. It's the r you know, we when we would show those prototypes to people just depicting the idea, it was really obvious that

1:23:13 People cared more about or they cared a lot about getting understanding reality and what was going on, and they were willing to disagree with like their side, so to speak, to recognize that. Um And I think that's not always that obvious to people. Uh, the world does feel really polarized, but

1:23:30 Um, people definitely are willing to cross partisan boundaries to to get to you know, accurate information and that's why the product works. It feels like As we rely more and more on what We know and s and understand, but the world is becoming

1:23:45 Social media. Online and moving this quickly. Feels like It's like we're s it's I'm so thankful this exists. Because Otherwise

1:23:53 It just be what do we trust anymore? Like this is like It like This being out aligns with we need this thing to exist. At the same time. And it feels like

1:24:04 At the same time there's also this people just Like I just don't trust Yeah. I think people have shifted from I trust what I read to okay, I shouldn't just believe everything I'm reading. Uh is there anything there you're noticing about just how people

1:24:17 Uh think about news they see. And there's shift of just like I'm not gonna believe everything. Is there anything that you've noticed about just like Human behavior just the way it's not.

1:24:27 We've shifted understanding what is true. We haven't done any research, you know, should we look broadly at at how people's perceptions are are changing there, but I certainly have found in myself that particularly seeing notes. Um, I am more sceptical about what I read at first and I I think that's

1:24:46 been helpful and we hear that from people that they Think about things a bit more and Um yeah, I think that's a good Secondary Effect and benefit of a

1:24:57 of something like this, which is the more you see the patterns of how what you're reading can be wrong. The more you can thoughtfully question it. And Try to

1:25:08 get a better understanding of what's really going on. So it's like, you know, historically I think this was called like media literacy, but A basic idea of like Can you understand the ways in which things Can go wrong. Try to cut them yourself.

1:25:20 And an another aspect I think we help with that is discovery of the community notes. Uh I think Often, you know, if it before community notes, you could have just been living in a little news filter bubble, or maybe there were fact checks out there. That you should have been reading, but you weren't Discovering them, uh right.

1:25:38 So so the the fact that the note applies It is directly attached to the post. Um and visible by anyone who sees the post. Oops. you know, cross th those filter bubbles and and can kinda

1:25:50 I I think for some people it's the first time they've they've actually seen counter arguments to, you know, uh to to claims made uh in their own. A little echo chamber. That's incredible. Yeah. I I love the point you're making about how it actually teaches people to s be a little more sceptical of the things they read. Like it's a education. uh system more than e more than just here's this one thing is wrong.

1:26:11 Mm-hmm. I love that. Okay, just a few more questions. Uh There was an audience question asked on Twitter. We all asked on Twitter, what do people wanna know about community notes? One was actually why you guys switched to anonymous contributors. What was the decision between behind that?

1:26:26 Yeah, that was you know, we we had this pilot um where we were testing the w with a small number of contributors, like thous a few thousand contributors, and we learned a lot through that pilot. Probably the biggest thing we learned was related to Anonymity or pseud pseudonymity of contributors. We had originally assumed that it was important that people contribute under their like real handle or their real name or whatever it was. We actually like the first prototypes Depicted that We kinda thought that would be important for like people trusting the note.

1:26:55 And actually was totally wrong. We were like it was the the rear the best option was actually opposite of what we first tried. Um we found a few things. One People were hesitant to write a note on a controversial topic. Because they didn't want to get like attacked or harassed online.

1:27:12 And so Some people were comfortable doing it doing this, but others were not. And so it meant there was more potential good notes to be written. um then we're getting written. And this is very clear feedback from the pilot. Two And this is super interesting.

1:27:27 People are actually more willing to cross partisan boundaries. When they are anonymous or pseudonymous. than when they are under their real name. And it intuitively makes a lot of sense. Like if you publicly are using your name, you are you you feel are affiliated with one side versus the other.

1:27:45 You might hesitate to kinda like be perceived as breaking with that side. But You may actually You know

1:27:52 For example, find a note helpful that's critical of that side. And Yeah, um, there's a bunch of studies that show when people are anonymous, they're much more willing to cross bound partisan boundaries and work with the other side, agree with the other side. And and we saw that too. And so By allowing people to just to be pseudonymous, you actually get more honest answers about what they really think and it helps find disagreement. That really so counterintuitive. You know, you never hear the opposite always and it's so interesting when it's the opposite. Yeah. Yeah. I think this uh the the same principle applies to making the likes private. I was just thinking that. Yeah. Yeah. I like a lot more stuff that's a little

1:28:25 It's tough. I I wouldn't have liked. Yeah. It allows it allows freedom for honesty, which is pretty great. Um And and we you know one of the criticisms of pseudonymity is like, you know, it can generate Maybe like people have reached their their quality the quality threshold that they put out there. But we have so many quality mechanisms in the system that that wasn't an issue. So Um we could keep quality high while while opening up for that honesty.

1:28:48 Another question, you touched on this a little bit, which is Around navigating the existing trust and safety apparatus of Twitter. Which to as you described, basically previously it was like we make decisions on what

1:29:01 Yes. True and not And There's every company works this way. You guys basically upended that. Like here's a completely different way. You have no control over what is what we say is true or not.

1:29:13 Talk about just that experience of kind of overcoming that, I imagine very difficult curdle of like okay, forget all that. We're gonna do it totally different. Yeah, it was definitely what we were proposing was very different. I will say that I think people were sort of open minded to it, generally speaking. Um And

1:29:29 I think everyone had a sense that what was being done at the time wasn't really working that well or solving the problem and you know, people were open to new ideas. So that was that's like a good foundation. But I think one thing we did That was probably very helpful in that. Is we wanted the product to prove itself at any point. Like first it had to prove that

1:29:49 people could possibly f you know find notes helpful. Then it had to prove that people could possibly like write these notes that would be good quality. And so Any time that we were proposing Doing something with the product. Like

1:30:01 running some research test or running the pilot or expanding the pilot. We always Had the data that had convinced us that that was a good decision. Like we w we were Stepping

1:30:15 into the next phase of expansion that made sense. And so I think We probably rarely proposed anything that Seemed unwise. Um, like'cause we were holding such a high bar for quality ourselves. Um I think yeah, I s I suspect that.

1:30:31 Went a long way. So it's partly what I'm hearing is like uh take it step by step to prove this is actually working. And partly. be confident it is working you're to yourself before you try to convince, say, the trust and safety team this is the way to go. Exactly.

1:30:48 Uh is there any was there like a moment along that journey of just like, okay, here, this like it shifted from no way this is a thing to okay, wow, let's actually consider this. Or is it this very gradual process? Whether other people were saying no way to well, let's not. Yeah, just internally of just like okay, we're gonna actually come stop this trust and safety. w uh way of operating. And instead rely on community notes. Is there like a moment of like okay, let's actually make that switch? Or is that Elon actually? Is that the big switch?

1:31:15 The biggest change there happened in X, um The biggest changes prior to that were just the decision to put this out there. And uh you know, n and have it be operating at Yeah, in public at first US wide scale.

1:31:31 Um, but yeah, then the big the bigger switches came in the X period. I I think even even though there was, you know, original research uh you know before before Bird Watch had even started. Uh our community that's even started. From external researchers showing that You know,

1:31:47 Crowdsource fact checkers uh can do, you know, lay people can do about as well as fact checkers, and actually the agreement rates were kind of similar between the groups. Like I think even though that research was out there I think there were definitely a lot of people who didn't really believe it could work until it already worked. So Hm. Basically prove it. Prove that it works. And

1:32:06 Yeah, that makes sense. Versus just a bunch of docs and strategy and thinking it's just like look, it's actually working, you can see for yourself. Yeah. Make sense. Okay. Possibly last question.

1:32:17 Uh we'll see which fractals of of questions you guys bring up here. Uh I referenced this a couple of times this Uh Incredible achievement of keeping a project alive. Through

1:32:28 Jack. And then I have this no and and Kayvon running the show then, and then Parag running Twitter, and then Elon. And then Linda taking over CO. Uh quite rare, especially something this visible, this impactful to

1:32:42 Everything that X is Any lessons or Keys to that actually. working over this project surviving throughout.

1:32:50 So many work. Changes and leaders. It definitely has been a crazy time to be building something. Uh, it's been fun. Uh the craziness has been entertaining. Um

1:33:03 I think we know one One reason the perhaps the product has has done so well and survived is the product itself the nature of the product itself. It is Designed to produce information that is found helpful by people who normally disagree.

1:33:18 And so even if you have CEOs or leaders who might disagree, like there's a good chance actually they'll find it helpful. They'll be like, wow, this thing does produce pretty, you know, useful output. So I think there's something in the nature of the product itself that you know, when people see it, whatever side they're on, you know, left, right. what are up down, um, they're likely to find it pretty helpful. Um so I I do think that helps. Um Uh I also think the team executed really well.

1:33:41 Um We had ambitious goals that were exciting. They solved A real problem. This is a real problem that matters in the world. Uh

1:33:52 Uh at every step. You know, as as we talked about the product needed to prove itself and you know, we would make sure it proved itself and we would Bring the results that convinced us and we would share those with people. And so they would say, Oh yeah, it I agree, it kinda proved itself. Let's Take the next lead. You know, we've we've done that all along the way and we continue to operate that way.

1:34:13 And I think Like that focus on Da outcome and goal that Matters.

1:34:21 And executing against it. Really helps. Like The team did not get distracted. By much. All through the period during which the acquisition happened.

1:34:30 There's a lot of opportunity for distraction. This team was shipping like every week. Like we were super focused on the goal, like Let's Make. Let's do it.

1:34:41 Get these nose out there and Um yeah, I think people saw that execution and and were excited to support it. Yeah, like it's working. Why would we must at that and it's important and it Keeps us from having to hire tens of thousands of

1:34:55 Of people to fact check. You know interesting thing about that is No one ever asked us or brought up or seemed to care about anything related to cost savings. Um in this process. And I think that's like a Pro like a

1:35:09 Uh an assumption people have outside the company that like this must have been a reason there was interest in it. But like That was never a goal. There's ne not at all why the project was started. It's not why people were excited about the project. And I think that's also, you know, for people outside who maybe don't see the conversations, it's like kind of a heartening thing to know is that the focus was always on solving the problem. The other approaches, even if you had ten thousand

1:35:33 People doing it. Right. it the real issue is that they don't work that well because They're not trusted or they don't scale or they're too slow. And so Um

1:35:42 The goal was really always just It's like help people stay informed at scale. You know, it's building an internet internet scale solution to an internet scale problem that people like. Something I heard about you, Keith, when I was asking people about How this worked and why this worked so well.

1:35:56 is that they describe you with a uh having a very low ego. And that allowed you to give up this whole team and power the influence and just the name, forget it, whatever you want. We'll call it community. That's great. Is there anything there you can share of just like h how how you Think about that and how important that is as a product leader. To have a low ego.

1:36:18 Um you know for me this This project, I feel like I get to do community service with this project. Like my I like my I see my work is in service of the people in the community. And like that's what motivates me. The only thing that I care about is delivering The outcome

1:36:37 That the world finds helpful. And so in some ways the project has it's not been about like ego or results, right? It's about it's like about truth seeking, like let's find not truth in the f sense of like what information is true, but like let's find out what's actually gonna make this work. Um like how does it need to be how does it need to be structured.

1:36:57 What is it what should it be called? Like whatever is gonna produce the best outcome is what we should do. And Um, so I think You know, I feel more attached to the product being helpful than to anything else. And so, you know, to whatever degree might seem like low ego is probably more a result of wanting to actually solve the problem.

1:37:15 And I think I think partly what I'm hearing is just if you win and succeed, good things will happen. Yeah, focus on that. Certainly satisfying things will happen. It's very satisfying to have people appreciate it. Satisfying that like People on the left and right. You know

1:37:28 Love it. It's satisfying that even people have received notes, love notes, and screenshot them and post them like that's amazing. It feels It feels you know so good to have helped Give people that and um

1:37:40 And yeah. You know, it's very motivating. It's great reason to wake up in the morning. It's absurd this has work, but it's also like of course This would work. Of course something like this should work. It's like such a interesting.

1:37:51 It's of the internet. You know, that's why it works. Oh man. Where's community units coming from here? What should people look what's happening, where's it going, what's uh what's the future. We're always working on

1:38:05 You're basically more better notes faster. So we wanna there's clearly opportunity to get more notes out there. Um, we wanna the want them to stay as good or better than they are. We wanna get them there faster. So We're always working on like core product changes um to help deliver that. Like recently, for example, We um just released an update to what we call the community notes bat signal, or the ability to request a community note.

1:38:29 So we anyone on X can say, Hey, I think this post needs a community note and now they can even add a source. Explaining why so that when a prospective writer sees that It's much easier for them to write a note. Um so we're always working on on core things like that, core algorithm improvements.

1:38:47 Um I think there are also New frontiers that show a lot of potential. Um AI. And LMs are one. It's easy to imagine a lot of ways that um AI could assist

1:39:00 the people in this task they're doing of of trying to get information out there quickly. Um maybe You Jay should talk about the supernotes work that uh we've done with Some folks outside the company.

1:39:13 Yeah. Uh so I mean one cool thing about having Public data and code is that uh extra normal researchers can can collaborate with you, uh and In this case, um The supernotes of this idea that

1:39:27 We can uh Basically take existing noses input. uh existing proposed notes that are actually you know, maybe they have some problem, maybe they have the whole part of the story, maybe they're worded in kind of a biased way. I'm basically take all these in.

1:39:42 Have an LM generate a ton of different variants. And then basically make the simulated jury. Uh So uh to to basically You know

1:39:51 Get a representative group of contributors of for community notes who would be writing the note and try to predict based on their past ratings how they would rate these LM generated notes. And so this way you can actually Um Yeah, rather than just like having an L and write a note from scratch and hoping it's good.

1:40:09 You can kind of like simulate the entire committee that's rating process. Uh and and You know, explicitly create notes that are likely to be readed helpful by people. So I I think ideas like that um are are very promising uh for the future. Um And is it a nice way that that LMs and humans can work together.

1:40:28 I think like I obviously, you know, a agents can can browse the web. too and and and that's one way that the you know, you could imagine agents assisting humans as you know maybe Maybe checking whether

1:40:41 uh sources actually supported by the note. um or a note is actually supported by the source. Although then you get into things like Well Uh you know, are are people going to actually be as diligent yeah, right now I think raters are very diligent because they know it's just

1:40:55 Some community notes contributor wrote this, like I better check this before I read it uh read it helpful. But Yeah, uh hopefully people Um yeah, we can design things in a way such that uh people don't trust uh the output and actually verify it themselves before issuing a helpful rating. Yeah, that is such an interesting area.

1:41:13 To explore where you want to avoid AI hallucinating slop. P Slap uh versus make it easier and and scale it even further.

1:41:22 Interesting challenge. What's cool about this project in addition to this the AI element is that it's being done outside the company. Like we talked to earlier about like trans the open source transparency. Like the key reason we we made this all open source was so people could see how it worked. But the dream Is actually that it's

1:41:39 The it's not just that the contributions So the notes and ratings are from the people. But the dream is actually the product's built by the people. Like w what if the scoring algorithm were significantly or entirely written by the public? Like that would be incredible. And supernotes is

1:41:54 Probably the first very substantial potential change in like the algorithm of the way it works. That was coming kind of coming from the outside and plausibly could be part of the core. Um so We'd love to see the product go in that direction as well. Sweet. Go super nuts.

1:42:10 Well, guys, the work you're doing is tremendous. Uh I think uh This is every product person's dream, I think, to work on something like this. Small team. Uh lots of support, lots of impact, like just like innately interesting. And so I think this is gonna inspire a lot of people.

1:42:27 Uh so let me just ask you is there anything else you wanted to share? Anything else you think might be helpful for folks? To leave them with. Sure. I guess one thing that uh just I thought was interesting over the course of working on this product is just there's uh I think in in a similar way to how retweets originally were not something like Jack came up with. I think user just started doing it and then it became a core part of the product. There's a huge way already in which uh there's just a lot of surprising things that people wanted to use community notes for that I don't think we really expected.

1:42:56 And it's it's Kind of cool to see those uh, you know, uh user desires uh kind of emerge. I think like one example. You know, I I guess we'd have always been imagining um political uh type of misinformation, but um for whatever reason there's like um

1:43:16 You know, a lot of people who love debating met whether Messi or Ronaldo got more goals. Um I guess the uh it's kind of kind of a funny one. Uh the uh There's a community moderation aspect, right? So like I think we also thought that Yeah, this would be specifically for Um adding context to misleading or potentially misleading information.

1:43:36 But what you can see is that um there are some notes that um go beyond that um towards like uh calling out um Content that they think is spammy or something. Um So I I think that's that's just

1:43:51 I guess another just another dimension on which community notes is a a product that's like Driven by the people itself. That's so beautiful. Basically they're trying to keep Twitter or slash X uh healthy and they're just like, No, this should be taken down, this tweet of spam. Yeah.

1:44:06 I love that. Uh is there an answer on the messy versus who is the other Ronaldo. Ronaldo. Okay. Uh is there like a definitive fact there? It's com that's that just unknown unknowable. Yeah, I guess I guess that's an interesting one because it's Uh it's a case where Raiders are actually very polarized, uh and I I guess it actually kinda fits into the the core algorithm where there's some people are just diehard messy fans or Ronaldo fans just like they could be on politics.

1:44:32 Um So actually uh you know specifically model that topic. Uh as well as some other topics so we can Like That's the main people's opinion on on that particular debate. It's kinda kinda funny that something like that would emerge. You're saying that's the most controversial topic on X. Ronaldo versus Messi.

1:44:50 It's a controversial one. Oh wow. Who knew? Okay. Uh Keith, is there anything you wanted to add?

1:44:57 Uh yeah. Yeah, community notes Is cool itself, but I think what it points to about society is actually even bigger. Um Society often feels really polarized.

1:45:09 You hear people talk about it all the time, like no one can ever d agree on anything. But actually like community note shows you People really can agree on quite a lot, even on super controversial topics related to politics and everything. There's a lot of agreement. That's my notes work. And I think that's a really

1:45:25 Big reason for optimism about the world is it while it might feel Polarized, there's probably like an eighty percent You know, set of people. that agree on quite a lot of things. And imagine if we could use the same kind of approaches we use with notes, but to find

1:45:41 agreement on legislation or policies or things like that that People want the government of the world to do. Like possibly we could get a lot more momentum behind these ideas that the people really want and everyone would be a lot happier. Like maybe 10% of the people on the edges wouldn't be happy, but like I bet there's a lot of agreement that we are not identifying

1:46:01 And if we did it. We'd all be pretty happy. So I don't know. I think it's easy for people to feel pessimistic about the world, but I think this is uh this product is a good reason to be optimistic about the future. What an incredible way to end it.

1:46:14 Uh I can also see Keith white people want to join you and work with you and work on this team. Appreciate it. You do want to join. We are hiring an ML engineer. You get to work on these amazing problems with us and have a lot of fun. So We're accepting applications at x dot com slash community nets. Okay, great. I'm glad you give the roll.

1:46:34 Oh man, you're about to get flooded. Uh guys. Thank you so much for doing this. Um, is there anywhere other than that place to go off uh join the team as an ML ML engineer, is there any other place you want to point people to, either your socials or anything else? I'm Kate Coleman on X. Um

1:46:50 A please reach out if you have any feedback or wanna help us out, whether you're gonna wanna work here or wanna do something from the outside. We'd love to talk. Yeah. I'm uh Underscore J A Y Baxter underscore.

1:47:03 Alex, uh Yeah, I think in particular, you know, besides just using community notes Um, it it would be great to to get more substantial contributions, you know, I could pull requests, um, collaborate on projects like supernotes. Um, I think that's the most exciting type of stuff.

1:47:20 Um people do want to contribute. Ship some code, guys. Yeah. That's amazing. Guys, thank you uh so much for doing this. Thanks for having us, Money.

1:47:28 Thank you so much. Hi, everyone. Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also, please consider giving us a rating or leaving a review, as that really helps other listeners find the podcast. You can find all past episodes or learn more about the show.

1:47:50 at Lenny's podcast dot com. See you in the next episode.

What you see above is a preview of the first minutes. One unlock costs 10 credits and covers this episode forever: full segment and word-level timestamps on this page, plus .txt, .srt, .vtt and word-level JSON downloads, as many times as you like.