OpenAI CRO says stronger models will align models; safety researchers push back
trevposts · x · 2026-11-18
Safety researcher Nathan Calvin questioned OpenAI CRO Mark Chen's comments on an a16z podcast about recursive self-improvement and alignment. Chen said chain-of-thought monitoring is today's most effective alignment tool and "the future of alignment has to be stronger models aligning other models," with RSI "intricately and deeply in this loop." Calvin notes this sounds close to letting AIs align AIs — which OpenAI has denied is the plan — while CoT monitoring is already deteriorating and will degrade further. The episode also featured John Schulman on what RSI means, why research taste is hard to teach, and why AI may be alignment's best tool.
More from AGI Musings
- Claude constitution's 2023 vs 2026 wording: from rule lists to a empathy heuristic — teortaxesTex · 2026-11-18
- Investor claims media is fake fear-mongering AI outages to stoke hate — econoar · 2026-11-18
- We Are the Misaligned AI: an ML Practitioner's Evolution Analogy for Alignment — theteknosaur · 2026-11-18
- Noah Smith predicts 4.6% annualized per-capita US real GDP growth through 2060 — inductionheads · 2026-11-18
- Timothée Chalamet is reading Kevin Roose's 'The AGI Chronicles' — and lent it to Josh Safdie — kevinroose · 2026-11-18
- Why veteran teachers' boredom argument makes AI tutors education's holy grail — RachelVT42 · 2026-11-18