"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis

Scaling "Thinking": Gemini 2.5 Tech Lead Jack Rae on Reasoning, Long Context, & the Path to AGI

770 snips
Apr 5, 2025
Jack Rae, Principal Research Scientist at Google DeepMind, shares insights on the revolutionary Gemini 2.5 model. He discusses the surge in efficacy of reasoning techniques and what this means for AI's journey towards artificial general intelligence. Rae dives into the interplay between human data and model behavior, the ethical challenges of reinforcement learning, and the nuances of AI reasoning. Listeners gain a unique perspective on the future of AI, exploring its potential applications and the hurdles that lie ahead in achieving true cognitive capabilities.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

RLHF's Steady Progress

  • RLHF has been used to improve reasoning for some time.
  • It's steadily improving, and recent progress has crossed a threshold of noticeability.
INSIGHT

AI's Moore's Law

  • Model improvements often appear sudden to the public but follow a smoother internal trajectory.
  • These improvements sometimes feel like a "Moore's Law" for AI.
ADVICE

Research Allocation

  • Focus on known problems lacking existing research.
  • Use AI tools like Gemini for summarizing and querying research papers, especially with long contexts.
Get the Snipd Podcast app to discover more snips from this episode
Get the app