Epoch AI's JS Denain Debates RSI, US-China Gap, and AGI Timelines with Nathan Lambert

Interconnects (Nathan Lambert) · rss · 2026-09-22

Nathan Lambert's Interconnects podcast features an in-depth debate with JS Denain, who leads the Insights Team at Epoch AI, covering RSI predictions, the US-China AI gap, and jagged capabilities.

RSI evidence is thin: JS argues that published evidence from OpenAI/Anthropic on AI accelerating AI research (e.g., researcher Codex spending doubling monthly) does not warrant expecting a software intelligence explosion within six months, though it merits tracking.
Insider information: He is unsure whether labs see dramatic internal metrics the public doesn't; public evidence plus priors already justify attention to the dynamic.
X-risk: He finds Evan Hubinger-style views (≥10% x-risk over a decade) reasonable despite deep uncertainty, and argues the core disagreement across the field is really about how huge and how soon capabilities get.
Capabilities maximalism: Looks like the AI 2027 scenario — AI automating AI research, massive robotics progress, AI-run factories forming a self-sustaining economy with accelerated science.
Other chapters: how far behind Chinese models are, whether distillation explains the gap, what Chinese job postings reveal about their labs, open vs. closed model safety, how Epoch AI operates, and what a frontier post-training recipe looks like.

Original post →

Epoch AI's JS Denain Debates RSI, US-China Gap, and AGI Timelines with Nathan Lambert

More from AGI Musings