Debating RSI, the US-China Gap, and Jaggedness with JS Denain of Epoch AI

Newsletter/Podcast · Listen · (gathered 2026-09-23)

• The discussion between Nathan Lambert and JS Denain explores the highly uncertain trajectory of AI, debating the pace of AI self-acceleration (RSI), the US-China capabilities gap, and practical bottlenecks in areas like robotics and model safety.
• Denain leans towards “capabilities maximalism,” envisioning an “AI 2027 scenario” with rapid robotics and industrial growth, while Lambert is more skeptical of immediate 10x productivity for all AI researchers, distinguishing between efficiency gains and fundamental intelligence breakthroughs.
• Denain estimates the US-China AI capabilities gap at 6-8 months, updating to believe distillation (e.g., using frontier models for training data) is a significant factor in narrowing this gap, potentially increasing it to 8-9 months if distillation were stopped.
• Lambert expresses strong skepticism about rapid, widespread robotic industrialization (2-5 years), comparing its progress to self-driving cars. Both acknowledge the “leakiness” of frontier models and the inadequacy of current safeguards, expecting regular “Hugging Face-style” incidents rather than immediate catastrophic misuse.