
Lessons from Noam Brown
The items below are things Noam Brown said on podcasts about reasoning models, test-time compute and AI progress. — Noam Brown on Unsupervised Learning, Ep 49: OpenAI Researcher Noam Brown Unpacks the Full Release of o1 and the Path to AGI (December 2024).
On the Dwarkesh Podcast, Brown said: "The problem is that as you push that further and further, you hit a latency bottleneck". — Noam Brown on the Dwarkesh Podcast, Agent swarms, alignment, & recursive self-improvement (September 2026).
On the Dwarkesh Podcast, Brown said: "I suspect that something like writing a novel would be very unparallelizable". — Noam Brown on the Dwarkesh Podcast, Agent swarms, alignment, & recursive self-improvement (September 2026).
On the Dwarkesh Podcast, Brown said: "I think this explains a lot of why startups are able to disrupt incumbents". — Noam Brown on the Dwarkesh Podcast, Agent swarms, alignment, & recursive self-improvement (September 2026).
On the Dwarkesh Podcast, Brown said: "The other thing is that because these AIs are jagged and they're, they're exceptionally good at some things". — Noam Brown on the Dwarkesh Podcast, Agent swarms, alignment, & recursive self-improvement (September 2026).
On the Dwarkesh Podcast, Brown said: "Like most capabilities, that could be used for good things or bad things". — Noam Brown on the Dwarkesh Podcast, Agent swarms, alignment, & recursive self-improvement (September 2026).
On the Dwarkesh Podcast, Brown said: "There is like a possibility that we go in the other direction, that actually every generation of models, we're able to make more and more aligned". — Noam Brown on the Dwarkesh Podcast, Agent swarms, alignment, & recursive self-improvement (September 2026).
On the Dwarkesh Podcast, Brown said: "Like, if we had chain of thought monitoring on for those models, it would've just like immediately shut it down". — Noam Brown on the Dwarkesh Podcast, Agent swarms, alignment, & recursive self-improvement (September 2026).
On Latent Space, Brown said: "When you work on these games, you kinda have to understand the game well enough to, like, be able to debug your bot". — Noam Brown on Latent Space, Scaling Test Time Compute to Multi-Agent Civilizations (June 2025).
On Latent Space, Brown said: "It was one of the things we realized over the course of the project that, like, oh yeah, we, you really benefit a lot from just having, like, larger language models". — Noam Brown on Latent Space, Scaling Test Time Compute to Multi-Agent Civilizations (June 2025).
On Latent Space, Brown said: "Like, there are some questions that I think don't really benefit from System 2". — Noam Brown on Latent Space, Scaling Test Time Compute to Multi-Agent Civilizations (June 2025).
On Latent Space, Brown said: "In fact, the vast majority of resources were focused on scaling that up". — Noam Brown on Latent Space, Scaling Test Time Compute to Multi-Agent Civilizations (June 2025).
On Latent Space, Brown said: "You could fall into this trap of, like, well, I just want one model to pose really difficult questions and the other model to solve those questions". — Noam Brown on Latent Space, Scaling Test Time Compute to Multi-Agent Civilizations (June 2025).
On Unsupervised Learning, Brown said: "I think the right way to think about it is that every time you want to scale these models further, there's a cost to that". — Noam Brown on Unsupervised Learning, Ep 49: OpenAI Researcher Noam Brown Unpacks the Full Release of o1 and the Path to AGI (December 2024).
On Unsupervised Learning, Brown said: "The problem is that, OK, well, if you want to 10x it each time, then at some point that becomes an intractable cost". — Noam Brown on Unsupervised Learning, Ep 49: OpenAI Researcher Noam Brown Unpacks the Full Release of o1 and the Path to AGI (December 2024).
On Unsupervised Learning, Brown said: "It's like, no, actually, we can improve this paradigm further and make the scaling a lot better". — Noam Brown on Unsupervised Learning, Ep 49: OpenAI Researcher Noam Brown Unpacks the Full Release of o1 and the Path to AGI (December 2024).
On Unsupervised Learning, Brown said: "Maybe we should try some more things and get more and more signs of life". — Noam Brown on Unsupervised Learning, Ep 49: OpenAI Researcher Noam Brown Unpacks the Full Release of o1 and the Path to AGI (December 2024).
On Unsupervised Learning, Brown said: "We want to see it take a hard problem that involves many steps and break that problem down into smaller pieces that it can tackle one by one". — Noam Brown on Unsupervised Learning, Ep 49: OpenAI Researcher Noam Brown Unpacks the Full Release of o1 and the Path to AGI (December 2024).
On Unsupervised Learning, Brown said: "I think something like O1 is something that scales really well with more data and more compute". — Noam Brown on Unsupervised Learning, Ep 49: OpenAI Researcher Noam Brown Unpacks the Full Release of o1 and the Path to AGI (December 2024).
On Unsupervised Learning, Brown said: "If you don't have those resources, then it's hard to push the frontier forward". — Noam Brown on Unsupervised Learning, Ep 49: OpenAI Researcher Noam Brown Unpacks the Full Release of o1 and the Path to AGI (December 2024).
On Unsupervised Learning, Brown said: "I think that we're going to see a major shift towards inference compute, and if there are ways to optimize around inference compute, I think that's going to be a big win". — Noam Brown on Unsupervised Learning, Ep 49: OpenAI Researcher Noam Brown Unpacks the Full Release of o1 and the Path to AGI (December 2024).
On Unsupervised Learning, Brown said: "I was very skeptical about the state of things and the hype around the progress in AI". — Noam Brown on Unsupervised Learning, Ep 49: OpenAI Researcher Noam Brown Unpacks the Full Release of o1 and the Path to AGI (December 2024).