"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis Historic AI Developments & the Emerging Shape of Superintelligence, from the Consistently Candid Podcast
111 snips
Mar 8, 2025 
Sarah Hastings-Woodhouse, host of the Consistently Candid podcast, dives deep into recent AI advancements and emerging challenges. She highlights breakthroughs in reinforcement learning and the competitive race for AGI between the U.S. and China. The discussion touches on the implications of AI memory and reasoning, as well as the ethical considerations in aligning AI with human values. Hastings-Woodhouse also explores the potential for superintelligence and the complexities surrounding governance in a distributed AI landscape.
AI Snips
Chapters
Transcript
Episode notes
Reinforcement Learning Breakthrough
- Reinforcement learning on sufficiently powerful base language models works remarkably well, unlocking new reasoning capabilities.
- This could lead to the emergence of many unusual AIs due to reinforcement learning's tendency to produce strange behaviors.
Democratization of AI Training
- Training powerful general-purpose models and shaping them post-training is becoming easier and cheaper due to scalable inference compute.
- Distributed training reduces the need for massive data centers, enabling diverse groups to train large models, making governance harder.
Shifting Geopolitical Landscape
- US AI leaders have shifted their rhetoric, embracing an AI arms race with China.
- This race raises concerns about catastrophic outcomes and hinders trust-building efforts.