David Krueger compiles ML research on why superalignment difficulty remains an open problem
DavidSKrueger · x · 2026-10-27
David Krueger released a curated doc, 'Selected machine learning research on difficulty of superalignment,' arguing very little ML research actually seeks to establish when advanced ML systems are likely to misalign. He leans toward MIRI's view that safe superintelligence with current methods is hard-to-impossible, but says extreme confidence either way is unreasonable — even a 1% extinction chance is too high — and technical work is badly needed.
Related event: Cambridge's Krueger Curates ML Literature on Superalignment Difficulty(2 posts)→
More from Safety
- Elvis estate sues Anthropic over 274 songs as LSEG sells data for $3.535B — YvesMulkers · 2026-10-27
- If ETH is false, LLM productivity gains may be eaten by rising security costs — jfischoff · 2026-10-27
- OpenAI, Google, Anthropic Reportedly Building FINRA-Style AI Self-Regulator by 2027 — emmanuelvivier · 2026-10-27
- IBM Launches Verifiable Identity and Control for Enterprise AI Agents — emmanuelvivier · 2026-10-27
- Mathematicians urge AI labs to stop giving designated mathematicians early access to model results — AndrewCurran_ · 2026-10-27
- AuthorizationBench: LLM agents overstep user authorization in 19.2% of runs — rohanpaul_ai · 2026-10-27