Real Alpha
Watch modeSubscribed

Dwarkesh Podcast · 4 min · Nov 25, 2025

Self-play can trade compute for scarce training data

Ilya Sutskever says self-play can create learning without external data, but past competitive setups teach a narrow set of skills. He sees debate and verifier-based criticism as related approaches.

Ilya Sutskever

Ilya Sutskever is cofounder of SSI.

Co-founded Safe Superintelligence in 2024 and leads it as CEO, following earlier breakthroughs beginning with AlexNet and work at OpenAI.

  • Dwarkesh Patel, Host · Host, Dwarkesh Podcast

    Co-wrote The Scaling Era, an oral history of AI; named to TIME100 AI.

About this segment

Ilya Sutskever says self-play can create learning without external data, but past competitive setups teach a narrow set of skills. He sees debate and verifier-based criticism as related approaches.

From “Ilya Sutskever – We're moving from the age of scaling to the age of research”, Nov 25, 2025. Watch the full episode

“the reason why I thought self-play was interesting is because it offered a way to create models using compute only without data.”
Ilya Sutskeverat 1:30:37 · Dwarkesh Podcast1:30:37

Takeaways

  1. Self-play is attractive when data is the bottleneck.
  2. Competitive self-play mainly teaches certain social and strategic skills.
  3. Debate and adversarial verification are related approaches.

In this episode

Full episode
  1. 18:51AI has returned to an age of research6:34
  2. 56:02Showing powerful AI could change how people respond7:47
  3. 36:23Research prototypes can work without frontier-scale compute6:30
  4. 47:14AGI may be a learner, not a finished mind6:50

  5. 1:29:36Self-play can trade compute for scarce training data3:41

  6. 1:32:54Research taste helps distinguish a bad idea from a buggy experiment2:50Up next

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

  1. 1:29:49

    But how do we elicit meaningful diversity among AI? So I think just raising the temperature just results in gibberish. I think you want something more like different scientists have different prejudices or different ideas. How do you get that kind of diversity among AI agents? So the reason there has been no diversity, I believe, is because of pre-training. All the pre-trained models are the same, pretty much, because they pre-trained on the same data.

  2. 1:30:16

    Now, RL and post-training is where some differentiation starts to emerge because different people come up with different RL training. Yeah. And then I've heard you hint in the past about self-play as a way to either get data or match agents to other agents of equivalent intelligence to kick off learning. How should we think about why there's no public proposals of this kind of thinking working with LLMs?

  3. 1:30:46

    I would say there are two things to say. I would say that the reason why I thought self-play was interesting is because it offered a way to create models using compute only without data. And if you think that data is the ultimate bottleneck, then using compute only is very interesting. So that's what makes it interesting. now the the thing is that self-play at least the way it was done in the past when you have agents which somehow compete with each other it's only good for developing a certain set of skills it is too narrow it's only good for like negotiation conflict certain social skills strategizing that kind of stuff. And so if you care about those skills, then self-play will be useful. Now, actually, I think that self-play did find a home, but just in a different form, in a different form. So things like debate, prove a verifier, you have some kind of an LLM as a

  4. 1:31:54

    judge, which is also incentivized to find mistakes in your work. You could say this is not exactly self-play, but this is, you know, a related adversarial setup that people are doing, I believe. And really self-lay is an example of, is a special case of more general competition between agents. The natural response to competition is to try to be different. And so if you were to put multiple agents and you tell them, you all need to work on some problem, and you're an agent, and you're inspecting what everyone else is working, you're going to say, well, if they're already taking this approach, it's not clear I should pursue it, I should pursue something differentiated.

  5. 1:32:32

    And so I think that something like this could also create an incentive for a diversity of approaches. Yeah. Final question. What is research taste? You're obviously the person in the world who is considered to have the best taste in doing research in AI. You were the co-author on many of the biggest, the biggest things that have happened in the history of deep learning from AlexNet to GPT-3 to so on.