TED Talks Daily AI won't plateau — if we give it time to think | Noam Brown
118 snips
Feb 1, 2025 Noam Brown, a leading research scientist at OpenAI, shares insights on transforming AI development by emphasizing slow, deliberate reasoning. He discusses the evolution of poker AI, detailing how thoughtful deliberation improved its performance against human players. Brown highlights the importance of patience and investment in AI to avoid stagnation, arguing that advancements are still on the horizon. He also illustrates how giving AI more thinking time can unlock extraordinary potential, challenging the idea that AI's progress is plateauing.
AI Snips
Chapters
Books
Transcript
Episode notes
Scaling Paradigm and Its Limits
- AI progress has been driven by scaling data and compute, exemplified by the transformer architecture.
- This scaling, however, poses financial limitations, raising concerns about potential plateaus.
Poker AI vs. Humans
- Noam Brown's poker-playing AI lost to human experts despite superior training data.
- This sparked an investigation into the role of deliberate thinking, similar to Kahneman's System 2.
System 2 Thinking's Impact
- Incorporating "System 2 thinking" significantly improved the poker AI's performance.
- A mere 20 seconds of thinking was equivalent to a 100,000x increase in training scale.
