Apr 25, 2023 · 1h 0m · no-priors
No Priors Ep. 1 | With Noam Brown, Research Scientist at Meta
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In the debut episode of No Priors, Meta research scientist Noam Brown discusses his breakthroughs in game theory AI—from solving poker to mastering human negotiation in Diplomacy—and explains why scaling inference-time compute is the key to artificial general intelligence.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 18% of the talking time here. How this is scored →
The hosts as informed peer 5.1 Guest teaching 4.6 Guest disagreement 1.4 The hosts pushing back 1.7
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 29:32 Noam counters Elad's claim on generality by pointing to sample efficiency
When Elad argues that the standard for AI generality is set unrealistically higher than human competence, Noam counters that the fundamental gap is sample efficiency.
Hardest push from the hosts ▶ 29:07 Elad challenges the human generality premiseElad directly pushes back against Noam's framing of human versatility, pointing out that individuals are rarely good at everything and questioning if AI benchmarks demand too much.
Biggest teaching moment ▶ 37:47 Noam calculates the 100,000x scaling penalty of omitting searchNoam breaks down the quantitative reality of planning algorithms, proving that matching MCTS performance without test-time compute would require an unrealistic 100,000x model scaling.
The host holds their own ▶ 39:20 Elad cites neurobiological modularity to question monolithic general architecturesElad demonstrates strong independent knowledge by referencing brain modularity, ablation studies, and evolutionary local maxima to interrogate the prevailing pursuit of single general architectures.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Noam Brown's Journey into Algorithmic Game Theory | 4 | 3 | 1 | 1 | Elad opens with a detailed question contextualizing prompt-based LLMs versus game-theoretic agents. Noam amicably shares his background moving from finance and economics into algorithmic game theory. The conversation is entirely cordial and biographical. | |
| AlphaGo, Deep Blue, and the Limits of Scaling Search | 5 | 4 | 1 | 1 | Sarah prompts Noam to explain search spaces in Go versus chess, sharing her own background as a former Go player. Noam outlines the difference in handcrafted evaluation functions versus pattern matching. The exchange is cooperative and educational. | |
| Tackling Diplomacy and Passing Undetected Among Humans | 4 | 5 | 1 | 1 | Elad asks what guided the selection of Diplomacy as a benchmark and what surprising outcomes emerged. Noam explains how Cicero operated undetected across 40 games by taking advantage of human assumptions in natural language. The tone is informative and collaborative. | |
| Cicero's Empathetic Dialogue and the Death of the Turing Test | 5 | 6 | 2 | 2 | Elad asks whether data bottlenecks will limit AI agent development. Noam gently pushes back on the conventional data-scarcity worry, pointing out that inference-time compute and reasoning architectures are the true bottlenecks. The dialogue is intellectually rich without hostility. | |
| Training Cicero with Human Data and Cooperative Self-Play | 5 | 6 | 1 | 1 | Sarah asks about training Cicero using self-play and limited human data from webdiplomacy.net. Noam delivers an in-depth explanation of why pure self-play creates alien conventions and how human behavioral priors are necessary in mixed-motive games. Sarah synthesizes the implications for real-world agent interactions. | |
| The Next Frontier: Moving Beyond Recreational Games to General AI | 6 | 5 | 3 | 4 | Elad actively challenges Noam's assertion that generality is humans' main remaining advantage, arguing that human domain specialization means the AI generality bar is artificially inflated. Noam clarifies that the key differentiator is sample efficiency rather than broad multi-task performance. This marks the most intellectually sharp exchange of the episode. | |
| Evaluating AI in Financial Markets and Real-World Negotiations | 5 | 4 | 1 | 2 | Elad inquires about AI applications in financial markets and programmatic smart contracts, linking it back to inference compute. Noam candidly notes his limitations regarding proprietary finance setups while explaining the challenges of non-stationary environments. Sarah then explores automated commercial negotiations. | |
| General Reasoning, Theorem Proving, and Inference-Time Thinking | 5 | 5 | 1 | 1 | Noam highlights the necessity of inference-time search over purely scaled neural networks, citing the 100,000x compute equivalence in AlphaGo. Sarah adds context regarding the Riemann hypothesis and code generation scope. The dynamic is respectful with mutual technical understanding. | |
| Architectural Generality vs Specialized Modules and Research Risk-Taking | 7 | 3 | 2 | 3 | Elad demonstrates substantial domain knowledge by citing neurobiology, brain ablation cases, and evolutionary local maxima to question whether general architectures should be preferred over modular sub-models. Noam acknowledges the validity of modular systems while clarifying his focus on general outcome capabilities. Sarah closes the segment highlighting high-risk research strategy. | |
| The Mechanics of Diplomacy and the Evolution of Centaur Play | 5 | 4 | 2 | 2 | Noam provides a rules breakdown of Diplomacy, explaining its unique private negotiation dynamics. Sarah and Elad discuss the durability of human-AI centaur play, with Noam playfully noting that in pure chess AI has already made the human partner redundant. | |
| Cracking Poker AI: Abstraction, Real-Time Search, and Nash Equilibrium | 5 | 6 | 1 | 1 | Noam details his PhD breakthrough with Libratus and Pluribus, explaining hand abstraction, real-time depth-limited search, overbets, and Nash equilibrium exploitation. Sarah notes how this upended traditional poker psychology. The hosts provide engaged, admiring wrap-up commentary. |
Statements from this episode (27)
Computer science iterates faster than economics because building requires no permission
“If you come up with an idea, you have to get it passed through legislation and it's a very long process. Computer science is much more exciting in that way because you can just build something. You don't really need permission to do it.”
AI was widely considered a dead field when Noam Brown began studying
“The idea of AGI was really science fiction. There were some people that were, you know serious about it, but very few, the majority opinion was that AI was, if anything, it was kind of a dead field.”
IBM's Deep Blue proved that scaling search works in AI
“We learned that scale really does work. And in that case, it wasn't scaling, you know, training and neural nets, it was scaling search.”
A five-person team could solve Settlers of Catan in a year
“To then, like, go to a game like Sotos of Catan, it just felt, like, too easy. Like, you could just take a team of five people, spend a year on that, and you'd have it cracked.”
Meta's Cicero played 40 Diplomacy games without detection as a bot
“But surprisingly, we managed to go, like, the full 40 games without being detected as a bot.”
People assume weird online text is human error before suspecting AI
“If somebody's saying something a little weird, because the bot does say weird things every once in a while, their first instinct is not going to be like, oh, I'm talking to a bot. Their first instinct is going to be like, oh, this person is like dumb or distra…”
The Turing test is no longer a useful measure for AI
“I think the Turing test is no longer really a useful measure the way it was intended to be. Certainly Just because we have bots that can, I wouldn't say they can pass the Turing test, but I mean, like they're getting close enough that it's no longer that usefu…”
Data availability is not the true bottleneck for AI scaling
“It's not clear that data really is the bottleneck on performance here. And I've talked to AI researchers about this, and I think there isn't as much of a worry about this as people might think. Probably that's because there's a lot more data that's out there t…”
A $500 million AI model will likely be trained by 2025
“You can probably easily 10 X that, you know, I wouldn't be surprised if there's a five hundred million dollar model that's trained in the next year or two.”
Inference-time compute is the missing scaling dimension for AI reasoning
“This is why I'm interested in the reasoning direction, because I think there's this whole other dimension. That people are not scaling right now, which is the amount of compute at inference time.”
Supervised learning on human games fails to produce expert players
“Like we also found in chess and go, we actually ran this experiment. If you do. Just pure supervised learning on a giant data set of human chess and go games. The bot that you get out from that is not an expert chess or go player. Even if it's like conditioned…”
Self-play negotiation bots trained without human data invent unintelligible languages
“If you train that bot from scratch with no human data it's going to, it could learn to negotiate, but it could learn to negotiate in a language that's not English. It could learn to negotiate in some like gibberish robot language. And then when you stick it in…”
An AI-generated novel rivaling Harry Potter could arrive by 2028
“I don't think you can get an AI to output like the next Harry Potter just yet. That might not be that far off. Maybe it's like five years away or something. But I don't think it's happening just yet.”
Humans require orders of magnitude less data than AI to achieve mastery
“Like how many games does it take for an AI, for a human to become a good chess player or a good diplomacy player or a good artist? The answer is orders of magnitude less than it takes for an AI.”
Reinforcement learning struggles in trading because financial markets are non-stationary
“I think the major challenge with Using things like reinforcement learning for trading is that it's a non-stationary environment. So you can have all this historical data, but it's not a stationary system and it's gonna like the markets respond to world events,…”
AI models could likely beat humans today in constrained business negotiations
“I think if you were to look at constrained domains certain negotiation tasks, I think that AIs could probably do better than humans in that today. I mean, I'm trying to think of like specific examples, but things like you know, if you wanted to negotiate over …”
AlphaGo's raw neural network performs substantially below top human players
“If you take out the planning that's being done in AlphaGo and just use the raw Policy network, the raw neural network, it's actually substantially below top human performance.”
Monte Carlo Tree Search fails in imperfect-information games like poker
“And that planning algorithm that's used in AlphaGo, Monte Carlo Tree Search, is very domain specific. I think people don't appreciate just how domain specific it is because it works in chess, it works in Go, and these have been like the classic domains that pe…”
An AI model could prove the Riemann hypothesis by 2028
“You know, it doesn't seem crazy to me that you could have a model that can prove the Riemann hypothesis within the next five years. If you can solve the reasoning problem in a truly general way.”
Next-token prediction will not replace big-company software engineers
“Like next, next token prediction is going to, is getting you surprisingly far. But I don't think it's gonna get you all the way there to like replacing, you know engineers at big companies.”
Cicero is the first major game AI breakthrough involving cooperation
“What's really interesting about diplomacy, aside from just the natural language component, is that it really is the first major game AI breakthrough in a game that involves cooperation.”
AI will eventually make human partners marginal in centaur Diplomacy play
“Eventually I'd imagine that these systems become so strong that, like, it kind of goes the way of chess, where, like, the human's just kind of, like, adding a marginal difference at the end.”
Inference-time search improved Noam Brown's poker AI performance by 100,000x
“If we were to add this search, this planning algorithm that would come up with a better strategy when it's actually in the hand, how much better could it do? And the answer was it improved the performance by about a 100,000 X. It was the equivalent of scaling …”
Noam Brown's six-player poker bot cost under $150 to train
“We did another competition that bought one and that bought Cost under a 150 dollars to train if you were to run it on like a cloud computing service.”
The multiplayer poker AI breakthrough was algorithms, not just scaling compute
“This wasn't just a matter of scaling compute. It really was an algorithmic breakthrough, and this kind of result would have been doable 20 years ago if people knew the approach to dig.”
All professional poker players now use AI bots for training
“And I should also say like the way professional poker players train now, They all use bots to assist them. It's a lot like chess where you play the game and then you have a bot analyze your play at the afterwards and see like, okay, did you make mistakes? Wher…”