Combining LLMs with AlphaGo-style deep reinforcement learning has been a holy grail for many leading AI labs, and with o1 (aka Strawberry) we are seeing the most general merging of the two modes to date. o1 is admittedly better at math than essay writing, but it has already achieved SOTA on a number of math, coding and reasoning benchmarks.

Deep RL legend and now OpenAI researcher Noam Brown and teammates Ilge Akkaya and Hunter Lightman discuss the ah-ha moments on the way to the release of o1, how it uses chains of thought and backtracking to think through problems, the discovery of strong test-time compute scaling laws and what to expect as the model gets better.

Hosted by: Sonya Huang and Pat Grady, Sequoia Capital

Mentioned in this episode:

Learning to Reason with LLMs: Technical report accompanying the launch of OpenAI o1.
Generator verifier gap: Concept Noam explains in terms of what kinds of problems benefit from more inference-time compute.
Agent57: Outperforming the human Atari benchmark, 2020 paper where DeepMind demonstrated “the first deep reinforcement learning agent to obtain a score that is above the human baseline on all 57 Atari 2600 games.”
Move 37: Pivotal move in AlphaGo’s second game against Lee Sedol where it made a move so surprising that Sedol thought it must be a mistake, and only later discovered he had lost the game to a superhuman move.
IOI competition: OpenAI entered o1 into the International Olympiad in Informatics and received a Silver Medal.
System 1, System 2: The thesis if Danial Khaneman’s pivotal book of behavioral economics, Thinking, Fast and Slow, that positied two distinct modes of thought, with System 1 being fast and instinctive and System 2 being slow and rational.
AlphaZero: The predecessor to AlphaGo which learned a variety of games completely from scratch through self-play. Interestingly, self-play doesn’t seem to have a role in o1.
Solving Rubik’s Cube with a robot hand: Early OpenAI robotics paper that Ilge Akkaya worked on.
The Last Question: Science fiction story by Isaac Asimov with interesting parallels to scaling inference-time compute.
Strawberry: Why?
O1-mini: A smaller, more efficient version of 1 for applications that require reasoning without broad world knowledge.

00:00 - Introduction

01:33 - Conviction in o1

04:24 - How o1 works

05:04 - What is reasoning?

07:02 - Lessons from gameplay

09:14 - Generation vs verification

10:31 - What is surprising about o1 so far

11:37 - The trough of disillusionment

14:03 - Applying deep RL

14:45 - o1’s AlphaGo moment?

17:38 - A-ha moments

21:10 - Why is o1 good at STEM?

24:10 - Capabilities vs usefulness

25:29 - Defining AGI

26:13 - The importance of reasoning

28:39 - Chain of thought

30:41 - Implication of inference-time scaling laws

35:10 - Bottlenecks to scaling test-time compute

38:46 - Biggest misunderstanding about o1?

41:13 - o1-mini

42:15 - How should founders think about o1?

The Perfect Crime