Artifact, the personalized news reader built by Instagram’s co-founders, is now open to the public, no sign-up required. Last month, Instagram’s creators Kevin Systrom and Mike Krieger, unveiled their latest venture as an invite-only experience, promising their news app would later evolve to include social elements, like being able to discuss the news with friends. With today’s launch, Artifact is dropping its waitlist and phone number requirements, introducing the app’s first social feature and adding feedback controls to better personalize the news reading experience, among other changes.
I’ve had access to Artifact since the beta launched a few weeks ago, and John Gruber and I discussed our initial impressions on Dithering. We were both a bit underwhelmed, although, as I note in the interview below, I have found the app increasingly useful over the last week in particular. That is on one hand a good thing — it suggests that Artifact’s core premise, which is that it learns your preferences — is a valid one; it also raises the question as to just how the new company will get users to stick around long enough to become useful.
I had the chance to discuss this question with Systrom and Krieger, ask what it was like being (very famous) second-time founders, how Artifact fits in the evolution of social media, and what the app might look like in the future. We also touched on Instagram’s beginning, the surprising benefits Systrom and Krieger gained from the acquisition, and the lessons they learned from Facebook that apply to Artifact, and more.
To listen to this interview as a podcast, click the link at the top of this email to add Stratechery to your podcast player.
On to the interview:
An Interview with Kevin Systrom and Mike Krieger About Artifact
This interview is lightly edited for clarity.
The Second Startup | Artifact | Customer Acquisition and Social Networking | Ranking and Machine Learning | TikTok and the Evolution of Social | Revisiting the Facebook Acquisition | The Future of Artifact and Content
The Second Startup
Kevin Systrom and Mike Krieger, welcome to Stratechery.
Kevin Systrom: Thanks for having us.
Mike Krieger: Great to be here.
So this is certainly a sort of founder/startup interview. Indeed, it’s the earliest stage set of co-founders that I’ve ever talked to, so it would seem like an obvious place for my usual opening questions about your background, “What did you do previously?”, things along those lines. But I think we’re all pretty familiar with your story! I guess that’s why I’m talking to you so early, it sort of makes the point. What’s it been like trying to build something new as the founders of Instagram? You’re not like a random set of co-founders, you are the Instagram founders, how’s that been particularly compared to the first time?
KS: Super easy, right, Mike?
MK: (laughing) Oh yeah. I mean, the things that get easier the second time around, the mistakes that you get to skip and the smoothing function you can apply to the difficulties of the startup journey where somebody once said, “It’s never as good as you think it is, it’s never as bad as you think it is” and you kind of can intellectually believe that the first time around, but you have to have the lived experience of having done it for a few years to really internalize that.
But the hard stuff is still hard. Finding the right product, finding the right partners, finding the right team — I think it gives you a bit of a leg up, but it’s by no means a sort of cheat code skip to level 10. I think we expected it and we got into it with full eyes open that it wasn’t going to just be you don’t pick up where you left off on Instagram, you’ve learned everything from the first time around but you’ve got to build it back up from scratch.
Is there something to where it’s harder because now you have the pressure because you’re the Instagram guys?
KS: I say to Mike, you can go to all these bookstores or these conferences and get advice about what to do the first time and what not to do the first time, but there’s literally zero books or zero conferences that focus on what not to do the second time and it turns out you just make a whole host of different mistakes. But to look back on our experience, I think that what helps having founded Instagram, is you fully expect the volatility of the experience, it’s not surprising at all. You fully expect to feel elated when something starts working, you fully expect to feel eviscerated whether someone quits or a major customer leaves — these moments, these crucible moments in your company where you feel like it’s all going to fail. We talk a lot about the second time around we just have this low-pass filter on, it’s all the stuff that it peaks above and goes below, it starts to feel less crazy and it doesn’t make it any easier per se, but you know how to deal with it. That’s been helpful.
Have you had those critical moments already?
KS: Yeah, I mean we started the company, Mike, what, basically two years ago? I mean, in earnest really a year ago, meaning when we finally got a bunch of people and started working on stuff, but I mean machine learning — I’m not sure how many of your listeners have direct experience with this, but the ones that do I think will commiserate — machine learning is a little bit like a dark art. It’s one of those things that doesn’t work until it works and there’s rarely an in-between. There are some days where we would wake up and just be like, “I don’t know that we can recommend articles to people. It seems like it should work, but I don’t know that this is going to work.” Then there are other days deep into the experience where we’d wake up and Mike would say, “I opened up my feed this morning and I’ve never seen such a reflection of me on a screen and this is totally going to work.” But I mean, I feel like I speak for both Mike and me in saying that it never gets easier the second time around, but you know what to expect.
MK: For sure there’s also an aspect of we’ve done Instagram, we’re coming out with a second thing, that’s a privilege. We’re talking to you and we’ve only been out for three weeks. You’re probably not talking to very-many-if-any three-week-old startups unless they really blow up like IG did the first time around. But for sure, I think we’ve always been our harshest critic, but there’s also this aspect of whatever we launch is going to get viewed and it’s not going to get viewed necessarily as, “Oh cool, the Instagram founders are starting something new and here’s their brand new app, they just launched v1.0.”, there’s an expectation of polish and craft and all of those things.
Kevin and I had talked, there’s pieces of the app that have felt done for six months that we could have launched six months ago and a combination of our own perfectionism and desire to get it right, but also probably carrying the question of, “All right, this is the second thing, we’ve got to get it right from the beginning” or at least have it feel more baked than we would’ve for an Instagram V1.
Mike, you mentioned the mistakes aspect before. Are there two collections? There are mistakes that you make with every startup and you are making them again, you’re avoiding some, I guess what are some of the ones you avoided is another way to put it, but to go back to your point, Kevin, there’s no book for second-time founders. What are the second-time founder mistakes that are new and original and that you’re getting to experience for the first time?
MK: With Instagram, there was the two of us until we launched and we finally hired a third person that week because we needed somebody to help run community ops and all of that and we hired our first engineer maybe a month after that. That led to a lot of insanity and sleepless nights but we moved quickly and it was the two of us and we could make decisions quickly.
When we started Artifact, we were conscious that we could have started and gone out and hired a team of twenty if we wanted to. There’s at least twenty great people I’ve worked with through the Instagram journey and we’re like, “No, we’re not going to go, we’re not going to be that extreme. We’re going to hire fewer than that. Maybe we’ll do seven to ten.” And honestly, even that was too many for starting it because so much of that early discovery period back when the app is just a sketch and you’re just getting week-to-week you’re sort of finessing from the highest level what the idea is going to be. It’s a bumpy journey and it needs to be, you can’t get too attached to one thing already, I think you’ll funnel down too quickly.
You remembered the pain of only being two people and maybe forgot some of the benefits.
MK: Yeah, and our significant others were like, “Well you’re definitely not launching this thing, just the two of you guys. We have families now and there’s infrastructure to take care of.” But I’d say that there’s two categories of mistakes and we’ve learned from them in that as well. One is I think you still need to start small. Seven was too many, maybe it should have been four. The other one is not everybody who is an incredible engineer that does great at an Instagram where the cycles are longer, even if I thought we moved pretty quickly at Instagram, there’s just a different pace, but there’s a lot more certainty on the underlying product. It’s not the shifting sands of a very early stage company. It just takes not necessarily a not necessarily different kind of person, but at least a person ready for a different kind of challenge.
KS: I’d say that the biggest challenge the second time around is not making things perfect before you launch them. Mike, you spoke to this, but I’d say if you ask all of my friends, if you ask my coworkers from Instagram, “What is the number one trait you would describe Kevin as”? I think perfectionist comes somewhere up in the top three. That doesn’t mean everything that I ever put out is perfect, it just means that I want it to be as perfect as possible.
What we realized with this company in particular is that it’s machine learning, so its main sustenance is data. Unless it has data, it will not work. It can’t exist in a vacuum, it’s not like Instagram where filters work from day one, regardless of the number of users. Our filters this time only work when you have hundreds of thousands of people using it and the machine learning can do its job. We had this moment where we had to decide do we go out with something — I’d say we’re proud of what we built, but we know it’s not where we want to end. There are a lot of things about it we wish for different. Do you do that given our personalities? That’s a really hard moment. That’s not something that we naturally want to do, but we realize there is no counterfactual here. There’s no world in which in which we stay private with 500 beta testers, most of whom are our closest friends where this thing gets otherworldly good. I mean ChatGPT needs to crawl the entire written Internet to get to where it is today. And even then they complain about running out of text, they want more data.
So the second time around, I think this perfectionism leads us to just be in a very different mode because the first time around we had no idea Instagram was going to be Instagram, we just launched. Sure, we were perfectionists about speed and we wanted to optimize here and there, but literally not a single thought process going on between the two of us, “What are people going to think”? Because we didn’t care, no one cared. So this time around that was harder but I’d say I think we did the right thing by getting it out. Probably six months too late, but we’re moving now and that’s all that matters. Every single day I wake up, there’s more data and the machine learning gets better and I can see it blossoming and that makes me really excited for the future of what we’re going to build.
Well I’m kind of interested in that from a product perspective because I have to say, the more I use Artifact, the more I find it useful. I thought it was not really useful in the first week at all to be honest.
KS: Of course. Yeah, of course.
Over the last couple weeks I’m like, “Wow, I’m actually finding stuff that I wouldn’t have found otherwise”, which obviously is the value proposition. So I have a couple questions on that. Number one, how much of that is the function of learning me versus as you just mentioned it actually is the entire corpus of people using it? So now when there’s thousands or tens of thousands or however many there are using it, what’s the function of improvement there? Is it the mass audience or is it me personally?
KS: I’ll touch on this a little bit. From a machine learning standpoint, the idea of cold start is as old as the domain of machine learning. The idea is, okay, you have someone who shows up, how good can you be and how quickly can you be good for them? The problem is, Ben, when you show up, you select a handful of interests. I’m not sure what you selected, but you selected some of them. We probably know you’re in a specific location, meaning general locale, we don’t know exactly where you are, but general locale. So the question really is how quickly can you go from basically blank slate default, which you said wasn’t useful, and I’d agree in general, to more and more useful? Mike has been using this product for, I don’t know, a year and a half, maybe longer, and we literally know every nuance of what Mike’s into. Mike really loves F1 car racing and he’s super into it.
Oh me too! That’s where I found the most links that I’ve shared.
KS: It will figure out the drivers he cares about, the teams he cares about, the drama he cares about and it will reliably serve him that. For me, it knows I love Japanese architecture, which is a very weird thing. Meaning it’s not weird, but it’s a very specific interest, it’s not like you’re going to sign up for something and tap Japanese architecture is your interest. That takes probably on the order of 100 to 1,000 reads, meaning taps into articles to start to learn that level of nuance. The interesting part here is that with lots of data, we can group you into an archetype of a person pretty quickly because it turns out, and I was telling Mike this the other day, the most fascinating thing about running this company so far and really working on the data personally, is that people like to think they’re very unique people with very unique consumption experience or tastes, and it turns out there are probably literally thousands of Ben’s out there and I speak in general.
It’s the Mike bucket, we just established that. I’m in the Mike Krieger bucket, just drop Ben Thompson in there.
KS: I’m speaking in very general terms, but in general there are archetypes of people and the thing is if you can pinpoint what archetype Ben is quickly enough, you can actually start to serve them things that feel personalized even more quickly. And they say, “Wait, how do you know that”? It turns out, well, we just know if you like interior design, you probably like architecture and you probably like recipes and cooking because we have lots of people that are exactly like that. If we’re wrong you probably write it off because you’re like, “Whatever, they’re serving me a recipe, who cares?” and we figure out from that mistake pretty quickly. So the answer — just to be very, very direct — is it’s both. It’s more data from you, but I bet you we have almost no data on you so far and then also it means lots of people signing up so that we can kind of understand these vague notions of a type of person so that we can serve more targeted things more quickly. This is all stuff that’s been used.
It feels like a sculpture approach. Instead of forming a persona, you quickly sort them into a persona and then chip away the parts that actually aren’t right and that’s sort of a shortcut to get there more quickly.
KS: Well said, and I’m probably going to steal that from now on, I appreciate it.
Call yourself Michelangelo!
KS: (laughing) It’s good.
Customer Acquisition and Social Networking
How much of a challenge do you feel in this “Getting to usefulness” quickly? I’m sure you had a whole bunch of signups particularly at the beginning and it’s very easy to churn out very, very quickly, particularly if that experience is not great. I know you’ve done stuff like streaks — which it’s hilarious how good this gamification works because it’s the stupidest idea ever that I have a streak of reading news articles, and yet I keep clicking that stupid notification to keep my streak alive. Has that been a challenge that you’ve seen to date is just people try it out, then they churn out super quickly, how do you get them coming back?
MK: One thing that’s been interesting, and this is as a second-time founder and having gotten to see Facebook that does user growth and analytics, I think better than anybody, we got to see it firsthand and apply the best of that to Instagram, is trying to calibrate what of that should you bring to bear when you have a beta group of 500 to 1,000 people? Are you just kidding yourself that you’re not going to get statistical significant data at all and what conceptually still matters? So even though we were onboarding, what was it like fifty to one hundred people for a while just through TestFlight, pre-launch-
KS: Per week.
MK: We were still being pretty analytical around what are their cohort curves look like? For the people that sign up do they stick around? What brings them back? What’s their dropoff point? Is there a clearing point where you’ve read enough that it knows you? And a couple things that stand out that I think kind of speak to this sort of “How good do we have to be on the first one”? Obviously we’d like to be as good as possible and the more data we get the better we can get on that first run, but assume we’re not going to be perfect because we’re not.
I think a couple things that really help — one is that Artifact isn’t a product where if your friends aren’t on it’s just not interesting and if they never join, it’s never going to be interesting. There’s a lot of great content out there that’s getting produced every day and we find people that even who might kick the tires, site it’s not working for them yet, do come back later. Sometimes it’s a week, some days it’s months. Sometimes we’re like, “Wow, this person showed back up.” It’s not like we learned anything more about them in the meantime, but we learned enough about people like them that the product got better or we added some new sources that were useful.
Two is push notifications, which we try to be thoughtful and smart about what we actually send to people, but that’s a great other way in which people like maybe a week later like, “Oh, that actually is a really good story that I want to click on”, and “Hey, the feed got better for these different reasons”. So hopefully we get people in and there’s always a group that’s going to stick. Then what’s kind of fun about Artifact that’s a bit different than Instagram is we got these other touch points over time that might be a trigger point for somebody checking it out again. Maybe we have some new announcement where they’ll bring it back in and I think as long as people give us enough tries over time I’m confident we keep getting better.
What are your social plans? Casey Newton mentioned in your launch interview that there would be the ability to comment on links and have direct messages conversations. As far as I can tell, it hasn’t been turned on broadly to date. There’s obviously an Instagram connection here: Chris Dixon wrote that famous post about Instagram, “Come for the tools, stay for the social network.” Is there a similar strategy here? It certainly seems like this could be a backdoor attempt to take on Twitter, which is a fairly insane goal for lots of reasons. Or is this all about algorithm and recommendations with a little bit of conversation tacked on? Obviously, you talked about the cold start problem with social networks, this is a great way to have content from day one, but is this the end state or does this evolve into a social network?
KS: Mike likes to give me crap sometimes because I spend a lot of time talking about what we’re going to be like five years from now, and I think it’s important to kind of draw the line between what we are now and as second-time founders understanding that you always start with a kernel, and then you let that thing blossom over time, but that we’re just kind of crazy enough to think that we will be around five years from now, so we can have a five-year plan.
Where we started is really, it’s the beachhead. It’s what can we be good at enough, with enough people and provide unique value that they’ll sign up and they’ll use it, and we’ll retain some portion of them? Mike was talking about retention curves. What’s crazy about the scientific method is you do all these experiments at a small scale and then you have tons of people sign up, and the curves look identical and you go, “Okay, well we did all these experiments when we were small, and now it scales to when it’s larger and now we keep these people and they start finding value.” Then they hang out and they start to use it and their data makes their feeds better, it makes other people’s feeds better. Then you can say to them, “Hey, wouldn’t it be great if X,” X being this additional utility that makes the experience even better, but it would’ve felt completely empty and suboptimal if you had had it from day one.
I think the mistake too many entrepreneurs make is they start companies assuming they have scale. We’ve seen no scale, we’ve seen hyperscale, and what we realized was what we have to do is we have to start with that utility and the best networks. I didn’t realize Chris wrote that post, but I should maybe in the back of my mind that’s what I was quoting and maybe I had read it, but I think all social networks start off as a utility and blossom into social networks long term, and that’s because they earn the right to be a social network, not because they start as a social network. That’s kind of where I think we are, is this utility phase where we’re trying to be as good as possible for our beginning users and we’re going to see as we grow what we’re allowed to add based on how many users we get.
Yeah, that certainly makes sense. One thing as far as the main social feature that I’ve encountered that I thought was interesting, and it’s not really a social feature, but is when you share a link to Artifact, and I swear I almost had a stroke when this happened and someone clicks that link—
KS: This is Mike’s feature, I’m going to blame Mike.
You get a notification that someone opened your link and the first time I was horrified, I’m like, “I don’t want that to happen” and I quickly realized that, no, actually I love it, that’s amazing. I want to know if people click my link and it was like I was just observing myself with a very clear demonstration of a revealed preference versus stated preference. If you had asked me if I wanted anyone to know that when someone clicked a link, I would say, “No way. That’s creepy, that’s weird.” Turns out I actually really like it.
Ranking and Machine Learning
And I’m interested in this in the context of, you go back to Instagram and Kevin, you mentioned in your interview with Casey that one of the lessons you learned was the way people responded to the algorithmic timeline. You go back when that was introduced, I was super pro and I’m like, “Of course people like it better”, but there was a lot of outrage at the time. I’m curious — were you in favor of that from when it actually launched? Both of you can weigh in on this, was there a lot of trepidation in Instagram? “Look, we’ve always been a certain way, if you ask anyone they says they want want it the same way, but we’re going to actually make this change and oh it turns out it’s better”, how has your thinking shifted as far as you have to give people what they don’t know they want, even if it seems a little weird or discomforting at first?
KS: Do you remember we actually added machine learning and ranking not to feed first, but where do we add it? We added it to the Explore page, right?
MK: Explore, yeah.
KS: We were sitting in that meeting, I specifically remember this and I don’t know why I remember this, maybe because I also had this feeling about the iPhone when it came out and I was like, “iPhone, what’s this thing?” and I remember feeling so anti-iPhone when it came out and of course Mike and I end up writing one of the largest applications for the iPhone ever.
One of the definitive iPhone apps.
KS: Yeah. I remember thinking when the team was like, “We’re thinking of using machine learning to sort the Explore page,” I’m not even sure what they call it now, but basically the Explore page and I remember saying, “It just feels like that’s a bunch of hocus-pocus that won’t work. Or maybe it’ll work but you won’t really understand what it’s doing and you won’t fully understand the implications of it, so we should probably just keep it very simple.” I was so wrong and I only remember it because I was so wrong, but you asked about feed, Mike would probably give you his anecdote about feed. But on the Explore page I was very anti and then I think I became pro only once I saw what it could do. Not in terms of just usage metrics, but just the quality of what people were served compared to some of our heuristics before. But I’ve been wrong really badly twice on fundamental technology shifts, but then I ended up investing a lot in personally and that was definitely the second one.
MK: I’ll share a funny anecdote about the Explore experiment. Facebook has all these internal A/B testing tooling and we hooked into it and we ran our first machine learning on Explore experiment and we filed a bug report and I’m like, “Hey, your tool isn’t working, that’s not reporting results here.” And they said, “No, the results are just so strong that they’re literally off the charts. The little bars that show it literally is over 200%, you just should ship this yesterday.” The data looked really good.
Shortest A/B test ever.
MK: Shortest A/B test ever. I think the feed ranking one is interesting and I’m really happy we’re starting with a rank feed because people don’t come into the expectations of the glory days when it was chronological or, “Bring back Chrono.” I saw this really interesting post from somebody on Mastodon actually and they said, “Hey, I just woke up and I missed 200,” whatever they’re called toots, I don’t even know, and overnight “I wish there was a way in which I could catch up on all this stuff.” That was actually a really big driving kind of thing behind ranking feed and Instagram.
Just to jump in, this is a huge thing for me being in Taiwan. The reason why I was always super pro-algorithmic feed and I was so mad at Twitter for not having one for so long is because I was always coming online hours after everyone else came online, there was no way I could catch up to this stuff. I need to know what was going on, so maybe that that’s a situation where my personal situation made me so pro-algo feed in a way that folks that were just in the moment, weren’t.
MK: And it made you aware of something that was already happening. We used to say internally, “Your feed is already ranked, it’s just ranked by order by created at descending.” Who’s to say that that’s the correct ranking for actually getting to most of your friend’s content? Maybe to tie this to Artifact too, the thing that was really interesting with feed ranking, and I think we probably could have done a better job of explaining this publicly at the time, was the whole goal, initially especially, was like “Can we help you stay up to date on what your friends are doing on Instagram”? That was the core of when it made Instagram special, and it was getting totally lost in a chron-ranked feed, because your friend posted 8:00 PM and in between then and the next morning, the brands you follow, the influencers posted fifty times because they have a whole team to help them post, and that friend post was totally buried. So it helped then to be really anchored in what the goal was, which was like, “Let’s be able to have a balanced feed that shows you both friend and interest content.”
It’s the same with Artifact. I think Kevin and I like to talk about a lot with this is making sure that we remain aligned on what is valuable to people, like, “What are people trying to get out of their day on Artifact”? It is not just how many articles I tapped on, did you get value? Did you come back because you actually read a longer-form piece? Can we continue to be really laser-focused on what does it mean to have a long-term relationship with the product that’s really serving them over time.
KS: Can I say something on this, which is, I think there are a lot of things that feel either weird or extractive in social, that feel much better in this product. So let’s actually just start with streaks for a second. Why do we have streaks? Streaks in a social product can feel extractive like, “Oh, they just want you to open it up again so they can induce their daily active metrics and stuff.” Actually we added it because we realized that news is basically only useful if you create a habit where you’re opening it up and you actually click on stuff and you actually engage with it. Therefore, it can be better for you in time, and unless we create that habit early, it’ll be useless for you for the long term. We were like, “Okay, how do we get people to come back? Well, streaks is kind of funny.” And for some reason Mike and I, like the number of times I filed bugs on Mike saying, “Hey, I traveled and crossed time zones and I lost my streak. Can you reinstate it?” It was a running joke internally, Mike’s main job was becoming the streak resetter.
A bit of pain that I feel myself as well.
MK: If you lose your streak, let me know.
It’s managing Wordle when you’re flying across the date line, very difficult. Sorry, continue.
KS: Yep, very difficult. But the idea is do we want people to engage with this a lot? Yes. But wouldn’t it be great if people engaged with written content by thoughtful people a lot? Wouldn’t that be awesome? And that’s one of the cooler parts of building this company is that, at its extreme people are learning about the world and that’s awesome. Can we get people to do that more? And how do we get people to do that more? On feed ranking specifically, I think one of the issues that we had at Instagram and Facebook has and others do, is that we’re kind of telling you who your friend should be and that feels wrong. Whereas in this case we’re like, “Yeah, we get it, there’s no way you want to subscribe to the RSS feed of a very large publisher that publishes 3000 things a day. We can help sort that for you.” A lot of these things that we learned, I think in a different context, have become increasingly useful in this context because they actually solve a consumer problem, not just a company problem and a company metric.
Isn’t that kind of the dark side though of the points you made before? Which is if you’re building a social network, you have to focus on utility first, and then you get the social aspects and that’s a very edifying view of how you build a company — don’t build for scale too early, focus on the need as you have the capability to meet it. But in this case it’s like, “Oh, we’re just trying to give people quality writing from writers they admire.” But if you actually follow that first ladder, aren’t the same sort of challenges going to arise in the dark arts as it were?
KS: Well, can you give me an example? I’m trying to follow exactly.
Well you say right now you don’t want to feel extractive as far as deciding who your friends should be or whatever, right? But if Artifact goes in a direction where there is much more stronger social aspects and maybe one day there will be generative content, maybe you’re going to be in the same boat?
KS: 100%. But the nice thing is we’re second-time entrepreneurs that have seen it before, and we can kind of avoid some of the pitfalls. There are certain companies that have been taken over recently by people that don’t have the first-time experience in the social network and some crazy stuff happens, right? Because there’s just certain things that you learn along the way, and I think we’ll be able to apply that to the second time around.
In general, I think we are trying to build a company that serves the people that use it rather than serves the company that we founded. That sounds like a nice thing to say, so I mean, what I was saying basically at the end of the last segment was you need to keep us to that. But the nice thing about being second-time entrepreneurs is we can kind of do that because we don’t have to work. We want to work on this idea because we think it’s really fascinating, we think it can be great for people and the only problem that we’re having right now is that it takes some time for you to realize that. What’s cool is that you use it a little bit and you’re like, “Huh, this is kind of interesting.” “Huh, this is actually kind of learning me.” We earn our way to each subsequent chapter and it’s not going to make it any easier when we get to some of these larger chapters with social integrated, etc., but I will say that I think we’ve learned a lot of lessons on the way and it’ll just make it, not easier, but more manageable.
TikTok and the Evolution of Social
You mentioned in that other interview about the evolution of social and this idea that obviously a feed should be pulling across the entire network and your friend network is a bit of a constraint. The exact quote was, you called them, “Unconnected graphs that are learned rather than explicitly created”, it makes total sense to you.
This approach was obviously brought to the American context in terms of TikTok. As I’m sure you know, TikTok is the American version of Douyin, which is created by ByteDance, whose first app was TouTiao, a news app. I’m curious to what extent you actually looked at TouTiao, or is it just you just sort of go up the tree in one direction, down the tree in the other and this is just an obvious place, it’s just sort of in the opposite direction from TikTok relative to TouTiao and Douyin?
KS: Well, what a lot of people don’t realize is that it’s not even Douyin was the first thing, Musical.ly existed. I met with the Musical.ly team when they — where was I? — I was in Shanghai for a board meeting of all things. And I met with Musical.ly and I went into their offices and there were probably thirty kids, and I say kids because I think they were probably 18 to 25 in this room working on this thing and I thought Musical.ly was so interesting and I was working on Instagram at the time, so obviously I had a vested interest in learning about what they were doing.
I remember having a conversation back at Facebook and people were like, “Yeah, I’m not sure there’s anything there beyond this flash of engagement.” I think ByteDance also felt that way but they were like, “There’s nothing there yet. But if we just layer on what we’ve learned at TouTiao, it’ll become Douyin, it’ll become TikTok.” Man were they right and that’s a really cool evolution to see, because I do believe in this idea that you shouldn’t have to click, “Follow,” we should just learn that you want to follow. Mike, do you remember this idea early on at Instagram where after you followed, you could set an expiration on how long you wanted to follow that person for, we were going to force you to re-up it at some point. Do you remember this?
MK: Right, to prevent this ossification of networks that inevitably happens. For the longest time, I don’t know if they finally went back on this, but we really resisted showing whether you followed somebody, if they followed you back, because we didn’t want to create the follows just for that social. It happens anyway, right? Because eventually you’ll find out or you can go browse their list. But yeah, the idea of some nuance in that follow relationship.
KS: It should all just be learned. I think it fits nicely into this narrative that I now tell where we were in college, we saw Facebook and the friend model. I was at Odeo, and I left before they created Twitter but then I saw them invent basically the follow model and then we were at Instagram and we saw the follow model get disrupted by this learned followed model, whatever you want to call it, the interest model, we need a name for the thing. But why that doesn’t exist in every single product we use, blows my mind and that feels like a fundamental shift that we all have to make.
And yes, we looked at TouTiao, of course, it was enormously successful, it still is today. The nice thing is, I don’t know when TouTiao was founded, but machine learning has changed dramatically since the day that TouTiao was founded. So why doesn’t that exist in the US? I think they actually tried, they had something called TopBuzz for a while, it didn’t quite work for whatever reason, but why doesn’t something like that exist in the US? It feels like such an obvious idea that some people would just be like, “Eh, it’s kind of uninteresting because you’re not generating images from prompts or something”. And it’s like, “Okay, fair”, but at the same time, neither were photo filters, it turns out that was a pretty good business.
We’re pragmatists at heart, I think, Mike, which is we find consumer problems that just should be solved and we solve them really elegantly or we try to solve them decisively. I’m not claiming victory, we’re still early innings here, but I find no problem in solving pragmatic issues at all, I think those are the best issues to solve.
MK: I’ll add one thing on the Chinese market I think is interesting. People over-rotate on this super-app thing. Oh, just because there’s an app or two that does a lot in there. Actually the thing I find more interesting about the Chinese market is how they’ll sometimes have an entire very popular app built around a primitive that we just don’t have here. One is Q&A. They have apps that are about Q&A and that’s the main — maybe Quora is the closest thing here — but imagine an app that is as ML-powered as something like TouTiao, but all around Q&A. I think that’s something that excites both Kevin and me over the long term is this personalization layer can be plugged into a lot of different verticals and different primitives, and there’s interesting cross-pollination you can do over time as well. Interesting written content in medium to longer form, call them articles, feels like or felt like the right place to start. But it’s really interesting to see how that same idea around unconnected content in a feed, with some social layer, but without that being the only thing, that drives what you see and how you consume it, and how that ends up manifesting.
So two points that touch on what you guys just said, and maybe this is a chance for Kevin to flex his five-year vision sort of muscles. Actually looking back, I think this is like eight years or so ago, I wrote this Article back in 2015 saying that Facebook needed to fundamentally shift its direction because the friendship network-driven feed was fundamentally limited. In retrospect, and I wrote a half mea culpa in an article called Mistakes and Memes, my analysis was half right, I think I was right about that being limited but my prescription — and it’s always easier to do analysis than prescription I will freely admit — was I wrote in that 2015 Article that Facebook needed to rely more heavily on traditional publishers to develop that interest graph. I think what TikTok showed is that user-generated content is even better. Of course the hit rate on a per unit of content basis is much lower, but when you’re aggregating such an astronomical amount of content, the absolute value that you can surface is much, much higher.
So to that end, in the near term, was I wrong, because Artifact is focused on professionally produced content, or is that really just the cold start problem and the ideal long run is having user-generated content, and knowing that that’s going to be much better than the stuff you’re sort of collecting around the web at this point?
KS: What people don’t realize, I think, about TikTok is that it’s effectively the largest experimental platform on the planet, which is to say if you just show a video to just enough people, you can collect just enough statistically significant data that says whether or not this video is super interesting or not. What’s cool about that is that if you have an enormous attention base and you can just spend that budget on showing people, frankly, a lot of crap that doesn’t work very well, if you can just find ten things that do really, really well by showing it to just a small number of people, you can then promote those things to the next stage of people and the next stage. What that gets you is a feed full of amazing content that you would’ve never seen otherwise.
We are obviously doing that with news on a very small scale, but I mean our push notification system effectively works like this. We try it out, we see if it’s interesting to people, we see if they tap on it, we see if they then read it and if there’s an enough engagement and enough signal, you then promote that to the next stage. You do that over and over and over and over again and we have this thing called The Board. The Board is what pushes we’re sending now, it’s kind of like a horse race and they’re all going against each other. What’s fascinating is to see what ends up winning is not what we would’ve predicted at all and it’s not just clickbait to be clear, that’s not what wins. Wins are like, weren’t you telling me today, Mike, there was some article on Anker launching like a new USB-C Hub or something?
MK: Yeah, 12-in-1 USB-C hub.
I want that notification. I love Anker stuff. Not in the Mike bucket I guess!
KS: My point is I would’ve never predicted that push notification to do so well and it’s incredible. People are really passionate about those products, it turns out.
So the reason I’m telling this story is because you can apply that to just about anything. It doesn’t have to be just written content. You were asking about your specific prediction on whether or not it had to be professionally produced content versus user-generated content. Our aspiration eventually is to let anyone be a publisher and one of the reasons why you can’t be a publisher right now is because you can’t find an audience. If you’re lucky enough to find an audience, then you go on Substack. If you’re lucky enough to find an audience, maybe you graduate to your own hosted blog or maybe you get hired by one of these big publishing groups, but it’s really hard to find an audience. The thing we can do is find that audience for you and long term, I totally believe we’ll have professional content all the way down to user-generated content, because if someone’s passionate about a very specific subject like Anker products, they should be able to write about it and be able to find an audience immediately.
That’s what’s so exciting about what I think where we’re headed is it doesn’t have to be just articles from the most well-known publishers in the world and to be fair, those publishers are amazing and their content is amazing, but wouldn’t it also be great to find really niche subjects that could be covered by people who are passionate about it? I think it’s actually both and that’s the biggest non-answer that I could give, but I do believe it is both and you just need the experimental platform in order to get there. Otherwise, it’s not possible.
I mean it sounds like what we’ve been saying Twitter should have been for the last decade.
KS: I can’t answer directly.
(laughing) It’s okay, we can move if you want.
KS: No, no, no. Actually I want to talk about Twitter because I saw it in its infancy and I’ve seen so many products go through this, and I think it’s okay to say this because Jack’s not there and that the founding team’s not there anymore. I think they would agree that the products that win evolve most quickly and typically away from their core.
I remember we were having a conversation with some of their early folks at WhatsApp and they were really anti some idea, it might have been payments or something, I can’t remember, or it might have been something about encryption. I honestly don’t remember the exact conversation, but what I do remember is very clearly making this pattern in my head of, “Wow, every time I see a startup stick to their original, what I’ll call their sacred cow or their 140 characters, that never works out in the long run.” Our version was square images, our version was just photos, our version was Feed. But I think we got lucky enough, and I’ll say lucky because I think we went kicking and screaming against some of those sacred cows and we were lucky enough to evolve and people still loved us as we grew. But you have to change what you’re doing drastically almost every three years, to stay alive.
To the credit of Facebook, or Meta I guess, but Facebook the product, they’ve certainly tried this multiple times and I think generally for every failure, there was an equivalently large win and it kept them alive, I think a lot longer than they would’ve had they just stayed a college social network. It’s one of the lessons you have to learn as an entrepreneur, you’re evolving or you’re dying, so choose.
Revisiting the Facebook Acquisition
Just to touch on Facebook for a moment, wasn’t it kind of nice to not be independent? You talk about things like publishers and UGC content and creators, and obviously that raises questions around monetization. One of the things is you didn’t have to worry about that because Facebook had this great monetization engine. Of course there was work to integrate it with Instagram, I’m sure there were kicking and screaming matches and figuring out how that would work or not. But at the end of the day, you didn’t have to build that. Is that something where you look back on with happiness, like “I’m glad I didn’t have to go through that”, or this is always the eternal question, I think for you two in particular, is there some regret, “We could have built a standalone company”?
To that end, I think there are two things that are underrated about you being acquired. Number one is obviously the fact that Facebook built the monetization engine for you. But two, it does feel like, to me, that Instagram was able to evolve in a fundamentally different direction than it would have standalone in part because Facebook, the app could pick up a lot of the pieces. Everyone’s talked about Facebook having all this functionality and Groups and Events and all this stuff shoved into it, none of which Instagram had to have. Instagram has more and more stuff shoved into it today, but it feels like that was a much more delayed addition of features and it was done in a direction that maybe was a little bit more true to what Instagram was and you had the luxury of doing that because of Facebook. Does that fit with the way you look back on it, or is there a bit of “Well kind of wish we had done it on our own and here’s our chance to do it this time”?
MK: The thing that was cool about doing some of these things inside Facebook was like — it’s funny, there’s like a classic idea in computer science, which is the second-system syndrome, which is like you’re like, “Oh, I learned all these things about what went wrong the first time, if only I’d known that the first time, I’m going to rewrite it and it’s going to be perfect”. Rewrites are often problematic, take years, get caught up in a huge morass of new issues that you find. I feel like we got to have a little bit of the best of both worlds in that we were getting to build the first system, but next door to a very established second system.
I use feed ranking as an example, which was when we decided that Ranked Feed was the right direction for the product, naturally when we talked to the Feed team at Facebook we’re like, “Hey, does it make sense to plug in” and they were like “There’s so much Facebook-specific logic in here, it’s going to take you guys two years to integrate. Does it make sense?” Instead, we were able to spin something up and in three or four months, it actually was really fast, this awesome team in New York did it. But they were able to say, “All right, how is this problem solved? How could we do it differently? What do they wish they knew the first time that we can do it and implement it in this way?” There were enough of those that, and I’ll speak on the CTO and builder’s side, that we ended up getting to do the, if not zero to one, then 0.5 to one and get to skip some of the early mistakes but still make it interesting.
But then we could do things — we had a principal called Do the Simple Thing First and it’s funny to think about doing that when you have millions or even a billion users, but even our ad tech, we know we couldn’t integrate right at the beginning just because the systems were different. So v1 was manually insertion, not even programmatic selling with scheduled buys, we did that for a couple months and learned a bunch there and then we started doing a smaller set of ads that we could run through it and then we finally were full auction mode. So yeah, there was that element.
I love building, I love getting to see things from the beginning and see it through. That was, I think, one of the reasons why we were excited to do another company was getting to do that again. Investing, for example was not wholly satisfying to me, I don’t think to Kevin either, because there isn’t that sense of, well there’s a hard problem to solve. We can learn from how other people solved it, but we’re going to do it our own way and who knows if it’s going to work. But we have to try, we have to build it, it’s so deeply ingrained in us.
KS: I was having a conversation with an investor two days ago and that investor invested in both Instagram and WhatsApp and he made the joke, he was like — I always make the joke to, I think it was Jim, but basically the guy who invests in WhatsApp out of Sequoia — that “the two values [of Instagram and WhatsApp] should have been flipped because one ended up being enormously valuable and one, I don’t know, we’ll see if it’s enormously valuable.” The jury’s out, but it doesn’t matter because basically they got a bundle. And it’s like if you take the aggregate amount they spent, it turns out they got a net positive ROI.
Yeah, exactly. Spend $20 billion, you’re doing pretty well.
KS: Yeah, it’s all good. I will tell you from my perspective, one of the wonderful things about having sold our company is actually that you can leave it. I know that sounds really weird, but hear me out on this. There are a lot of these CEOs that I see at the cocktail parties or the conferences, whatever, and they’re just stuck in the company they founded fifteen years ago and they’re working on the same problems and they’re trying to reinvent the company, but it’s really, really hard and they don’t really want to be a public company CEO anymore, but they have to be because they own all these shares. Mike and I just kind of get to do new things. Our aesthetic and our brand is just totally different, our approaches are totally different, we just get to hit the reset button and go. Meanwhile we’re living great lives with wonderful families and things are exciting and we get these new challenges. So I don’t know, there’s definitely this rationalization going on, but at the same time, I don’t know many people that get to hit that reset button and try new things. So that’s been an exciting chapter for me and I don’t regret that ability at all.
To your specific question around whether Instagram got air cover from Facebook in whatever way, man, I feel like it’s been years, so I feel like we can talk about this, Mike, at this point, but just the internal struggle of, “Was Instagram successful because of Facebook or in spite of Facebook”? It’s like, of course if you ask the Instagram folks, it’s in spite of. If you ask the Facebook folks, they’re like, you could have never lived without us, and the truth is likely somewhere solidly in the middle.
KS: But everyone had their opinions and man, there were so many fights about it. I was so sick of these conversations because it’s like, who cares? What counterfactual world are you living in where we have to figure this out? Why don’t we just go win hard regardless of whose fault the winning is. Let’s just go win. So I felt like there were tons of wasted cycles on the blame game of who’s at fault for winning.
What I liked most about Instagram, the Instagram experience inside of Facebook, was that we skipped probably five, six, seven years of development because yes, there were lots of technical implementation details, we couldn’t just plug in Ranked Feed, we couldn’t just plug in the ads. But man, once we got those ads plugged in, it worked, and it worked really well. Then we would do things like, we started ads on Stories long before Facebook did. That started making a ton of money and then all of a sudden all these advertisers got excited about that format across the products so I don’t know whose fault it was for winning so hard as a team, but I’m glad we won because it went really, really well. The fact that we get to do something new now I think is really special as entrepreneurs.
I don’t think that’s a rationalization, I think that’s a phenomenal answer, this bit about you were able to leave. There’s an aspect of doing what you’re doing now is a lot more fun and interesting for a certain type of person than being a public company CEO and the fact that people get stuck is a good point.
The Future of Artifact and Content
What does success look like? Right now as far as I know, you’re still self-funded, is this just a, “Hey, we get to experiment as long as we want because we don’t have that pressure”? Is that actually a detriment where there isn’t a succeed-or-die mentality?
MK: We talk consistently about this being, and it’s going to sound funny, hopefully the last company we found in our lives, and not because we don’t want to do it again, but because it can be a platform for realizing all of what we want do in terms of taking what’s at the forefront of technology — I think you can break this down to what Kevin and I do — we’re not researchers, we’re not the people trying to build things that are twenty years out from the technology and you have to squint, “Ah, maybe that’s going to work”. We take what’s at the forefront, on the cusp, and try to bring it to bear with what I think is strong product sense and engineering chops, and practicality. That’s the universe.
Yeah. So you rode the iPhone wave. Now you’re going to ride the AI wave.
MK: Yeah. Maybe there’s an aspect of that which can sound like bandwagon-jumping, it’s not quite that. I think it involves the taste to understand, all right, what about this is begging for consumer products to be built and how do we do that really well? Not like, “Yeah, this technology exists. Let’s go solve a problem with it.” But based on everything that we know, what can we build that will solve a problem? And the cool thing about that is that that’s a template or almost like a recipe that I think we can follow for a long time, and do it with the right teams and do it together and evolve over time, and that’s the core of what we’re trying to do. In general I see this as a platform where this is what we’re working on right now, but the approach doesn’t change. It’s just the manifestation that might over time.
In terms of what the stakes are, I’ve mentioned earlier we’re our own harshest critics. I think we’re also our own harshest standard setters. Maybe as first-time entrepreneurs at Instagram, we’re like, “Wow, some people signed up the first day. This is cool.” Even if that’s all that happens, it’s great, we’ve seen success and it doesn’t mean that everything needs to look exactly like that. But we do have the joint and lifelong ambition that if we’re going to do something, it better be really meaningful and whatever. Meaning might shift. We did COVID projects together during the height of the first and second waves. And ambition or success, there wasn’t usage numbers, it was like, “Hey, is this useful to people that are making decisions?” so that might shift over time. But shared recipe plus believing that at least has a good fighting chance to be meaningful in its fullest realization. I’d say that’s the kind of platform that I hope this can provide.
One more bit since you mentioned that you’re riding the AI wave. In the time since Artifact came out, I got a chance to meet Sydney as I recounted on Stratechery, and it really pretty fundamentally changed my view on AI potential. This goes back, Kevin, to the second part of my earlier question, as far as the five-year view. Of course, there are utility functions to what we’ve seen with these chatbots, but it seems like the real endgame of social, and I actually wrote about this in the context of Instagram last summer, but Sydney really brought it back to bear — it’s not just recommending content from across the network but actually generating content.
In that respect, Artifact does seem different than the cutting edge of “social”, which is not really social at all. First, as I mentioned before, it’s usually professional content, not user-generated content and perhaps it sounds like that’s going to change in the long run. But second, it’s trying to be factual and true. And you have made comments about this that you want to have a point of view on things like that, instead of say interesting and conversational, which humans often are. And maybe your take is that’s fine, there’ll be lots of different applications of AI, but do you sometimes wonder at the end of the day, that because you’re in this attention space, just how much user time and attention is going to be devoted to things other than news or facts or whatever it is that Artifact is perceived as being now?
KS: I think people always want to know what’s going on in the world around them, and the way they learn about that is through various mediums throughout time. It can be through the printed newspaper, it can be through a magazine, it can be through a radio show, it can be through a podcast, it can be through a video on TikTok, it can be on cable television. You name it, they can learn about the world around them. But that job — we use jobs-to-be-done theory a lot in our work — that job-to-be-done about learning about the world around you and going deep on your interests will never go away and we just happen to do it with a specific medium, which is easy to do asynchronously when you’re sitting there. Whether you’re in bed or you’re waiting for a meeting, you don’t have to listen to it. You don’t have to have it on audio, you don’t have to have headphones in.
It turns out that a lot of people really like to read. I joke with Mike that there’s this earned secret that everyone thinks no one likes to read. Everyone’s like, “Oh, there’s this shift to video.” And it’s like, “Yes, video is growing in terms of time spent” but it turns out people really like to read. Now, do they read the full article? Not always. Do they skim it? Probably. But they do enjoy written content, especially when there are images in it and they can engage with the content. And we’ve been able to build a product that I think serves people in that specific job very well.
But going forward, it turns out that not every article has to be written by a person. And I’m very excited for the world where if you are really interested, we were talking about anchor products before, wouldn’t it be great if you just woke up and there were a custom article written for you? And to call it an article, I think is probably saying it’s a horseless carriage with the car, it’s less that it’s a custom article written for you, and it’s more that it’s custom generated text that is exactly what you’re looking for. It’s almost like this preemptive scan of the internet for content or knowledge that you might be interested in and hopefully there’s a world in which we can be both curators automatically so you could imagine generating an article or whatever you want to call this thing, this living entity, that points you to the most thoughtful analysis or point counterpoint on different issues that you’re following right now. You could do that with a generative model at some point, maybe even today, but you could also just generate pure text and information based on what’s out there on the Internet if you’re really interested in a very niche topic. I find that future so enthralling.
Right, because it’s like cutting out the middleman, right? Instead of trying to collect a bunch of almost right articles that are in your interest, wouldn’t it be great to pull everything out of them and just give you the one-page summary of what happened in F1 yesterday or whatever it might be? But this is the question for all this sort of stuff, what about the source publications? Right now the worst part of Artifact is all the ads in the articles. You can go to reader mode, but it’s under a menu probably because you don’t want to make publishers mad. What tension do you feel versus the long term source of your knowledge, versus how you reveal it?
KS: I’m watching all of these lawsuits closely. There’s the guy who publishes code and says, “Hey, wait. GitHub CoPilot, basically somewhere in there is my code, I swear.” And then you have artists looking at Dall-E thinking “Somewhere in there is my art”. I think that’s a real issue we’re going to have to grapple with. Not we, Artifact, we, society. And I don’t have all the answers, but it’s going to be interesting to watch it play out. In no shape or form do we wake up in the morning saying, “Our goal is to replace publishers”, because it turns out reported content from the front lines with interesting, thoughtful, unique analysis that you can’t get anywhere else is not something you can simply parrot out with a model and so I think that a lot of people are going to have job security for a very long time in the publishing industry because of that.
At the same time, I think there’s this really interesting opportunity to allow for people to publish more easily, more quickly on topics, that may not be economical today. It’s the simple Amazon thing. For the longest time, you could only buy bestsellers and all these mom-and-pop bookstores but all of a sudden it was economical to sell interesting — whether it’s fan fiction or a specific genre — I actually think we’ll see more production in the long run, human production, because we can find audiences with this system than we would’ve without this system. And I think that’s a bright future for publishing, not a scary one. But Mike, we’ve talked a lot about this, what do you think?
MK: I think I keep coming back and this is probably something I think may resonate with your own experience, Ben: I think a lot of the value of great reporting or great content online comes from the personality or the perspective. It’s the writers that I keep coming back to over and over and over again on a topic and sure, maybe GPT-4 is going to be able to mimic the ideas well enough that it’s hard to tease them apart. But in general, there are either sources of analysis or voices or perspectives that I think differentiate, and I think that is if not irreplaceable, just very hard to replace and just that has that human aspect to it. The more we can find those voices — and those voices could be writing for a major publication, they could have their own blog, they could be starting their own blog now — and I think that is the human element that I get most excited about. If we could make their lives easier, that would be awesome, if we can find them an audience, that would be great. I’ve already seen that happen even in our beta test in early days. I actually had no idea this publication existed and I really like it, it’s covering the war in Ukraine really closely in a way that I hadn’t heard of before. Sports analysis — I’ve discovered sports writers that I hadn’t been exposed to before and I keep coming back to.
So I put my vote or my belief firmly in the writers’ corner, regardless of exactly what medium they’re being positioned in. And yes, I think there’s a long tail on local, this has been tread before, and a long tail on interests where either factual reporting or summarization on things that aren’t economical to do from a human perspective, will be useful. I was reading, I think it’s Neal Stephenson’s, not his latest novel, it’s two novels ago, and there’s this concept of social media editors and the wealthy people get humans to edit their social media feeds for them. If you’re not as well off, you have to rely on a public algorithm, and it’s dystopian in that way.
But there’s just way too much content out there for any single person to consume or even summarize and that’s where I think that if I can be a little bit more succinct, connect people to interesting writers and have them fall in love with them and have them form the long-term relationship and especially in places where it’s economical for that writer to make a living and hopefully can make that easier over time. Then, there’s a lot of other things like just giving summarization or generational leg up, will just help people stay more informed on their long tail.
Yeah, obviously I’m biased because I think that will hopefully be the outcome as well. I will say there’s a whole middle layer here that is super annoying. The whole headlines that are questions and don’t tell you the answer and I get mad in Artifact because of course I want to click it but I don’t want to reward them in the algorithm and think that they got any sort of attention. So I’m like training a pet here or something along those lines, but that’s what happens, the Internet. The middle gets wiped out and this certainly seems a push to original content and finding that and surfacing it and perhaps rewriting it, but these intervening layers that give you the opportunity to have a cold start might not be very viable in the long run. I could end it there unless you have any final words, I appreciate you taking the time.
KS: I do have a thought on that because I was speaking with a guy who did the algorithm at Instagram. His nickname actually internally was literally “The Algorithm” and we talked a lot about recommender systems early on. One of the points he had is that the way these algorithms are trained, the objective function that you give these algorithms to optimize, ends up having this strange, almost tertiary effect that it forces content creators to start creating content that fits what the algorithm wants to optimize for.
It’s why in TikTok, at the end, have you ever seen these videos? I’m sure you have, where literally at the end they say, “And that’s why”, and then the beginning words of the video are like, “You always cook chicken in butter”, or whatever and it loops and it feels perfectly loopable and they do that because they need more repeat view so that the algorithm goes, “Oh, I have a winner here and I’m going to show it to more people.” So the way you train your algorithm ends up forcing people to create a certain type of content, and what we have to be very careful with, I think on this particular point, is to not reward behavior where you obfuscate the actual content of the article, aka clickbait, where people write more sensationalist headlines or more sensationalist articles.
If we get big enough, there’s a real consideration on how we train the algorithm and what objective function we get or give it, and what content gets produced for it. We need to be very mindful of that because what I’d rather have is a world where we have more original content, more interesting content, and less of this gaming-the-algorithm content and it’s a trap that every social network falls into. But again, we started this talk by talking about what it’s like to be a second-time founder. We’ve seen a few things and what’s nice about going through this a second time is we have that in mind when we’re building it so hopefully we can avoid some of those pitfalls.
Sounds good. Well, Kevin and Mike, I appreciate you coming on and taking the time. Thanks for coming on.
KS: Thank you very much.
MK: Thank you.
This Daily Update Interview is also available as a podcast. To receive it in your podcast player, visit Stratechery.
The Daily Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly.
Thanks for being a supporter, and have a great day!