Commentary on Christiano/Dwarkesh: training-pipeline automation is desirable and safety-helpful
sebkrier · x · 2026-11-18
Sharing early takes on Dwarkesh's interview with Paul Christiano, the author argues that "RSI" is a misleading initialism better understood as "model training pipeline automation." He sees further automation improving AI capabilities as broadly right, desirable, and even useful for safety. Assuming a slow take-off, much of it is manageable in practice—important risks and fast capability growth are expected, but he's more optimistic about the endogenous societal and technical response. He firmly rejects claims that we may be "six months away from AI systems that can overthrow the government" or that government overthrow is a likely outcome of ML R&D automation.
More from AGI Musings
- Field experiment: Google AI Overviews cut publisher clicks 39.8% with no UX gain — paulnovosad · 2026-11-18
- Academia Has the Strongest Anti-AI Norms, Yet Grad Students Use AI Anyway — CatAstro_Piyush · 2026-11-18
- "College and the 9-5 path is over" — AI era anxiety goes mainstream — rand_longevity · 2026-11-18
- Models Will Be Commoditized: Burkov Backs Efficient Open-Weight Models, Doubts Overpriced IPOs — marlene_zw · 2026-11-18
- "If AI brings intelligence and opportunity to billions, dying industries deserve no sympathy" — VraserX · 2026-11-18
- Formal verification meets agentic AI: neurosymbolic approach pitched as a path to trustworthy agents — luislamb · 2026-11-18