CEOInterviews.AI
Start App
Get an email when Misha Laskin gives a new interview
One email per new interview. Unsubscribe in one click.
Misha Laskin
Co-Founder and CEO, Reflection AI

Solve coding, solve AGI [Reflection.ai launch w/ CEO Misha Laskin]

📅 Mar 07, 2025 Latent Space 28 MIN 2819 VIEWS 32 SEGMENTS · 2 SPEAKERS
Misha Laskin of Reflection AI introduces their new research lab, focused on achieving AGI by solving coding. Misha and his team led development of Gemini, GPT-4, AlphaZero, and more foundational work in the AI space. Links: https://www.reflection.ai/ https://latent.space/ https://x.com/latentspacepod Timstamps thanks to VidSkipper AI Chrome Plugin 0:00 🎀 Introduction to Reflection AI β€’ Introduction to Reflection AI and their mission to build reliable super-intelligent systems. β€’ Michelle Asin from Reflection AI discusses their stealth mode and upcoming announcements. 0:46 πŸ€– The Core Techno...

Questions asked in this interview

11
  1. 4:07Is there anything specific in the last six to 12 months that kind of made you decide, okay, now is the time to do it, or is it just a matter of, you know, the team coming together at the right time and the market being ready?
  2. 7:59But why so confident that it will lead to superintelligence?
  3. 10:32How are you approaching your data, or like, what are your strong beliefs on your data?
  4. 12:03And there's a question of, do we, you know, just like you believe that there should be native browsers, a native user interface, is there a different language that, like, will actually be better for AI to code in?
  5. 13:21What do you think, like, a move 37 for a coding agent would look like?
  6. 14:44So what do you think the role of open source will be in that?
  7. 17:00Like, what is the shape of the products thinking that you have so far?
  8. 20:16Are you doing some kind of β€” what is the state of the art in code indexing these days?
  9. 22:01Is it valuable having a, you know, a cluster of 10,000 GPUs that are just handling, like, a billion-token context, you know, for one user?
  10. 22:52How would you, I guess, encourage people to think about comparing your products versus, like, a Devin, like a Poolside, like a Magic?
  11. 26:00What's kind of like the culture?
Host 0:02 ↗
Hey everyone, welcome to the Latent Space podcast. This is another lightning episode, and I'm joined by my co-host, Swyx, founder of... hey. And today we are very blessed to catch up with Misha Laskin from Reflection AI. Welcome.
Misha Laskin 0:16 ↗
Thanks for having me.
Host 0:18 ↗
So today we're sort of recording this ahead of the announcement, but you have some very exciting news here. I think you're basically coming out of stealth. So, yeah, I mean, if you want to sort of lead off with, like, I think the last that people might have heard of you was when you raised your round from Sequoia, and you did a little podcast with the Sequoia team. But now you're actually ready to have your landing page say more than, like, two sentences about what it is you're doing. So maybe you want to introduce the world to Reflection AI.
Misha Laskin 0:43 ↗
Yeah, definitely. And on the landing page, I think more than one sentence β€” I think we have kind of one highly undescriptive sentence that lives on there right now. So definitely have more to share than that. So I think, I mean, the main thing is that we're kind of ready to share the, you know, what it is that this company has been working on, what our kind of sort of research approach is, you know, what our product approach is, and the team that's assembled around this problem. So I think that, you know, maybe I'll just start off, say, you know, something short, and then we can take it from there. So I think that, you know, what we've come together to solve is maybe β€” this was controversial 10 years ago, but now not so much anymore. I think that there are many, multiple labs that are pursuing the problem of superintelligence, and this is kind of the mission of the company, our company as well. We're building reliable superintelligent autonomous systems. I don't think that it's so much the difference in, like, the pursuit that, you know, makes this unique, but it's more around the team around it and a rather opinionated approach to how one should build a superintelligent system. So first on, you know, the reason we decided to start this as a company is we kind of realized that two building blocks, like on the technology side and the research side, had come together that even made this problem possible. And the first one was reinforcement learning, which a lot of members of our team pioneered some of the largest advances in that, including DQN, networks, AlphaGo, AlphaZero, so forth. And what was special about reinforcement learning is that it's actually β€” it's already a blueprint to how to build superintelligence. Like, we've had super β€” we've lived with superintelligence systems, but they are very narrow. So AlphaGo was superintelligent at the game of Go, but couldn't really do much outside of that. It was then retrained on chess and so forth and became superintelligent in other board games. And so while we kind of β€” we've actually been living with these superintelligence systems, we know what they look like, but they just haven't been very practical yet. And the other core technology that a lot of the members of our team helped pioneer was large language models. So we pioneered projects like Google's PaLM, Gemini, ChatGPT. What happened with language models is that they are kind of the opposite. They're these really general systems that are useful, but require a user to drive most of the experience. Like, we kind of think about language models today as almost like cruise control, where it's helpful almost everywhere, but you're still kind of driving the AI and, you know, doing most of the work as a user. So we kind of thought these two ingredients had come together where you could combine them and build not just a narrow superintelligence, but a general one on the computer. And I think that our particular approach to it is where is β€” maybe most people, or I won't actually β€” won't speak for other people. I think that it's kind of better to, you know, different people have their different judgments. But I would say that rather than seeing autonomous coding or coding capability as one of many things that we think a superintelligence will do, our core belief is that if you solve this problem, you solve the autonomous coding problem and build a superintelligent coding agent, that that thing will lead to superintelligence properly. So that kind of encapsulates, I think, the kind of team, the core beliefs that we share with the company.
Host 4:07 ↗
And just to do a quick historical timeline, so you mentioned AlphaGo, OpenAI β€” I was also working on RL on games with Dota in the beginning, and then everybody just focused on language. And I think now the pendulum is shifting back to RL. Is there anything specific in the last six to 12 months that kind of made you decide, okay, now is the time to do it, or is it just a matter of, you know, the team coming together at the right time and the market being ready?
Misha Laskin 4:32 ↗
It was really that. So Ioannis and I led a lot of the work for post-training and kind of RLHF for Gemini, and Ioannis being my co-founder. And when we shipped Gemini 1.5, we just realized that the models β€” like, models that were basically at GPT-4 level or above were capable enough as kind of starting points to then post-train β€” or, I mean, it's β€” I wouldn't even call it post-training, just train this reinforcement learning. So it's really that. I don't know if you would have been able to do the same stuff with an earlier model, like with a GPT-3 or GPT-2, but GPT-4 had this kind of base of intelligence that you could actually go from there. And by the way, this actually happened in AlphaGo as well, and some of these systems previously, where before you did reinforcement learning, they first did imitation learning on human games, and it was important that the human playing ability was sufficiently high for you to be able to then bootstrap on top of that and then train with reinforcement learning. That is, if you would have trained AlphaGo on very weak human gameplay, then for that first project, reinforcement learning wouldn't have taken you as far. Of course, AlphaGo then figured out transition to AlphaZero, where it was trained without human data whatsoever. But I think that's where the analogy breaks. I think in the era of language models, I don't think you'll be able to train superintelligence from scratch.
Host 5:59 ↗
Yep. And then on the environment, you briefly mentioned the computer. It's the idea that, you know, you mentioned code, but you want to be very computer-use focused. You know, some people say the browser is kind of like the new OS anyway. I'm curious your thoughts on where you want the agent to live in.
Misha Laskin 6:14 ↗
Yeah, it's a really good question. I mean, if you start kind of an autonomous agent company today, you might pick from one of two categories, maybe some others, but I think broadly speaking that you could either say, I'm going to build browser agents or kind of computer-use kind of agents more broadly, or build coding agents. And I think that, or, you know, our belief as a company is that the correct wedge and the correct starting point to this entire problem is the coding agent, because it's already β€” you know, software engineering is already what I would call kind of ergonomic for a language model. That is to say, you have to design the problem that you're striving for to be compatible with sort of what is intuitive for the intelligence that's working on it. And for humans, what's intuitive to us is, you know, using a mouse and keyboard and kind of geospatial reasoning, because that's just how we evolved. But language models never evolved. They were trained on the internet, and because they were trained on the internet, their priors for what is intuitive to them is completely different from what's intuitive to us. And so a language model, for example, has no prior for mouse movement. It, you know, it really never seen that on the internet, but it has a really strong prior for code. So out of the two categories, say web browser use and coding, coding is the only one that is actually already ergonomic to what the language model is good at. And so it's much easier for you to amplify good behavior, whereas if you want to do a web browsing company or something like this, you have to go and collect this gigantic dataset of human mouse interactions and figure out how to clean it up, because I don't know if you ever tried capturing a human's mouse, but it's extremely noisy.
Host 7:51 ↗
Oh, you know, nothing a common filter cannot fix.
Misha Laskin 7:55 ↗
Yeah, yeah. So code is, you know, all you need.
Host 7:59 ↗
It sounds like. But why so confident that it will lead to superintelligence? You know, I think you are one of the most strongly opinionated on this that I've met. Because I guess, like, obviously, I think a lot of people understand that, you know, code can sort of augment the LM in a sort of, you know, like, system one being deterministic, system two being sort of more non-deterministic or sort of reasoning-focused thing that all of them can do. But I don't really understand that leap to superintelligence that you mentioned. Like, I feel like this term means something special to you that we haven't really dived into yet.
Misha Laskin 8:36 ↗
I think that, I mean, so for me, it just means it's just a system that will be able to do most of the work that we want to do on a computer, and not just kind of do it, but do it more creatively than we could, right? That it'll discover, similar to how in AlphaGo, move 37 was this kind of move that AlphaGo did that was more creative than what Lee Sedol, the Go champion, could imagine at the time. Similar things, I think, will happen in, you know, computer-based work, where, you know, I think it's concrete in software engineering, where you ask a model to optimize some code for you, and it comes back to you with a solution that you didn't even know was possible, but is amazing. So I think that's kind of the world that we're headed into. But the reason coding is so fundamental for this broader vision is a belief that since code is already kind of the ergonomic form factor for a language model, we believe that a lot of software that, you know, a lot of UIs are built for humans today, like the GUI, as keyboard was built for a person. And over the coming years, there'll be more kind of AI-friendly or language model-friendly UIs. And what's friendly to a language model is code. So the way a model will be doing work, not just for coding and software engineering, is by basically making function calls or using SDKs or like programmatic interfaces that have been developed for it. For example, I think the browser will fundamentally change. There'll probably be a user-facing browser that relies on the UIs that we use, but there will be, you know, more friendly language model interfaces for, you know, basically doing the same things that we do on a browser. You can kind of think of this already happening, and that like the backend for like the OpenAI search products or Perplexity or some of these other companies, you can think about the backend for that is largely programmatic. And so they've effectively built these kind of AI-first browsers for themselves that I think that is more broadly coming.
Host 10:32 ↗
Yeah, I think like abstractly, I think AI-first browsers are more or less search APIs, maybe with a little bit more extension that you can take on those things. One thing I wanted to sort of dive into also as well is just the comment that you made about sort of imitation learning. And, you know, I dug up an old comment that they had from 2023 about how ChatGPT interleaves code and text, and it was basically just imitation learning on contractor data. Is that what you're seeing? I think, like, as you mentioned, like, a lot of approaches typically end up with some version of self-play as well that scales up. How are you approaching your data, or like, what are your strong beliefs on your data?
Misha Laskin 11:12 ↗
Well, you kind of β€” yeah, you kind of have to have both. So it's really important to both get the right initial data mixture from supervised fine-tuning, however it is that you gather it, that feeds the correct behaviors that you want to elicit out of the model, and then you kind of amplify those behaviors with reinforcement learning. So the key thing to reinforcement learning is that it only works β€” I mean, it can potentially work otherwise, but practically it only works when the agent has interacted with a reward, right? It's received a positive reward for what it's done, maybe one out of 10 times, one out of 50 times. But if it's getting zero reward, then you don't really know what behavior to amplify. And so that's why it's really important in the kind of initial mixture to already be doing things out of the box that are somewhat sensible, even if they're unreliable.
Host 12:03 ↗
Yeah, good priors are very important. Yeah. So, like, and then I think one more thing that, like, ties to a previous conversation we've had with Bret Taylor from OpenAI, which was, like, a lot of our code right now is just in languages that happen to be nice for humans to write, primarily Python and JavaScript. And there's a question of, do we, you know, just like you believe that there should be native browsers, a native user interface, is there a different language that, like, will actually be better for AI to code in?
Misha Laskin 12:32 ↗
You know, that's actually β€” that's a really fascinating question. You would think so. You would think that there's some better abstraction. Again, like, coding languages are mostly made for humans. You'd think there's probably a better abstraction that's more AI-native. It's possible that something evolves that is different from what we have today, but it might also look somewhat similar because the things that language models understand best tend to be the kind of piece of code that represented on the internet. You know, maybe β€” I don't think a lot of people would be happy with this, but maybe the kind of fundamental language of AI becomes Python or something like this. Like, everything gets bundled in Python SDKs. It's, yeah, it's not β€” I'm not sure exactly where it will go, but it's probably going to be something close to what's already native to an LLM or some abstraction of that thing.

16 more exchanges in this transcript

Sign in free to read the rest of this interview. No card required.

Sign in to read the full transcript

Cite this transcript

APA, MLA, BibTeX
APA

Laskin, M. (2025, March 7). Solve coding, solve AGI [Reflection.ai launch w/ CEO Misha Laskin] [Interview transcript]. Latent Space. CEOInterviews.AI. https://ceointerviews.ai/interview/3071997/

MLA

Misha Laskin. "Solve coding, solve AGI [Reflection.ai launch w/ CEO Misha Laskin]." Latent Space, 7 Mar. 2025. Transcript, CEOInterviews.AI, https://ceointerviews.ai/interview/3071997/.

BibTeX
@misc{laskin2025_3071997,
  author       = {Misha Laskin},
  title        = {Solve coding, solve AGI [Reflection.ai launch w/ CEO Misha Laskin]},
  howpublished = {Interview transcript, Latent Space. CEOInterviews.AI},
  year         = {2025},
  month        = {mar},
  url          = {https://ceointerviews.ai/interview/3071997/},
  note         = {Speaker-attributed transcript with timestamps}
}