Lenny's Podcast: Product | Career | Growth
00:34:38 · Adam Mosseri
[00:34:38] S0: One of the most common misconceptions is actually in the opposite direction. I think people assume that there's a much more detailed semantic understanding of everybody's interests and preferences in the algorithm than there is. Most of what's really driven the progress in the world of recommenders over the last 5, 10 years have been these large embedding models and these other techniques that basically produce artifacts that cannot be read by people. They're not legible. They're giant vectors. It's like, sure, I can show you the vector, but it's just going to be a bunch of numbers in a seven-dimensional space. And so when we talk about, does the algorithm know something? Usually we think in these more semantic terms, it knows I like surfing. And it's like, "Eh." It doesn't. It just has this big-ass number that happens to correlate with surfing. That said, I think that is starting to change. I think that one of the things that LLMs are enabling is they can describe in words, English or whatever language you prefer, what some of those previously-illegible artifacts are at least proximate to, if not mean directly. So this is like ... The thing I haven't really ... I posted about this this week, this thing called Your Algorithm. Basically the idea is we take a look at all of the stuff that you've interacted with, and then all of that is in an embedding space. You can think of embedding space as a map. You map a bunch of videos into the same map, and so videos that are close or similar. And now, we can just have an LLM just be like, describe that part of the map. And it can be like, "Oh, that is deep pour over coffee snobbery." And that's kind of amazing.
00:36:29 · Lenny Rachitsky, Adam Mosseri
[00:36:29] S1: That is so cool. You can ask the LLM to look at these numbers, and extrapolate here's the topic that you're interested in. That's awesome. [00:36:36] S0: Yeah. Or look at the videos. [00:36:37] S1: And the videos.
00:36:37 · Lenny Rachitsky, Adam Mosseri
[00:36:37] S1: And the videos. [00:36:38] S0: And so the way you ... Both. You can also embed concepts into that same space. And so I mean, embeddings are really the underlying technology underneath LLMs. That's how all the whole thing works. And so what we do now is we let you "see your algorithm." You can see what topics we think you're interested in and you can adjust it, you can add and remove things. But the idea here, giving people some agency back in a world where these social media apps are getting taken over by recommendations. But we can't do a lot of other things yet, but we will be able to do. There's things that aren't topical that you might ask for, "I want more fun content. I want to see my friends more. I don't want to see my high school's friends' kids' photos." I don't know, we can come up with a ... "I don't want to see seven photos in a row, but I'm happy to see six photos in a row." Whatever your mind can come up with. So we have a lot of work to do, and so I'm excited about that. But I think a misconception historically is, until recently, we don't really know as much about you as you think. We're just like, "Oh, you liked these photos. These people also like those same photos, and they like these other photos, so you might like those other photos." I'm oversimplifying, but that's Kind of how it worked. Now, only now, are we actually getting as sophisticated as I think people have assumed we've been for many years.
00:38:06 · Lenny Rachitsky
[00:38:06] S1: That is really interesting. One that comes to mind is this transition everyone eventually goes through to this algorithmic, broad global feed. It always feels, like people think, "I just want to see chronologically everyone I know and follow, and that's going to be my favorite feed," and it continues to be proven wrong. No, you actually engage a lot more when it's this algorithmic feed of things we think you will love.
00:38:31 · Adam Mosseri, Lenny Rachitsky
[00:38:31] S0: Yeah, it's tough because ... I mean, I posted this week this thing about agency, and I just got destroyed in the comments, which is just part of the job. I get it. But there are a couple issues with the chronological feed. So one is ... And some of this is the tension between an individual's interests and what works when you scale it up. So if you do a pure chronological feed, the incentive for everybody is to just post as much as possible, because it will always be at the top of everyone who follows you's feed as soon as you post. So what ends up happening is that the feed gets overwhelmed with professional content, with usually large company content and publishers. The New York Times can pump out 50 things a day. Your best friend won't. You might get one thing a week from them. And so, your feed just gets taken over. So part of it is the incentives that emerge, because when you design these systems, it's almost like designing a city. You need to think about, "Okay, here's how the mechanics work. What are the incentives that arise? How are people going to act within those incentives? And then what happens?" And the other thing is sometimes the most interesting thing was just not the most recent thing. Recency is an important input into relevance, but it's not the only one. My sister got engaged last night, and she's in Germany. [00:39:57] S1: Congratulations. [00:39:58] S0: No, she didn't. If she did. She's married. She got married last year. That's why it was top of mind. [00:40:02] S1: All right.
00:40:02 · Lenny Rachitsky, Adam Mosseri
[00:40:02] S1: All right. [00:40:03] S0: But if she got engaged and I missed it because she lives in Europe and we're different time differences, do I really want to see a picture of my brother's po'boy sandwich, or do I want to see my sister's things first? So it's tough. It's tough. I'd love to figure out a way to find the right balance. I want to give people agency over the experience, but I think it needs to be in a way that creates a system that makes sense. Not just for us as a business, which matters. I'm not pretending that's not an issue, but also for the overall community, because we've done chronological by default, or you can make a default, and you see not only does usage go down, overall sentiment goes down. The individual who made that choice might be happy at the moment, but when you just get pummeled with stuff you're less interested in over the course of months, we ask, we run surveys and massive scales. We just se people start to become less and less satisfied with Instagram. [00:40:57] S1: Kind of along these lines, everybody asks you about this these days, AI and content and how that all impacts everything that's going on. I want to ask you something I haven't seen someone ask you. Is the rise of AI content a headwind or a tailwind for Instagram versus other platforms? Do you think this helps or hurts you guys?