Cambridge's David Krueger compiles ML research on why superalignment is hard
DavidSKrueger · x · 2026-10-27
David Krueger (Cambridge) has released a quick doc, "Selected machine learning research on difficulty of superalignment," compiling ML work relevant to why aligning superintelligent systems is hard.
His key observation: very little ML research actually seeks to establish clarity on the conditions under which advanced ML-based systems are likely to be misaligned or dangerous — it remains an open problem. MIRI and others in AI safety hold strong intuitions that safe superintelligence is hard-to-impossible, but the supporting empirical research is thin. The doc is publicly shared via Google Docs.
Related event: Cambridge's Krueger Curates ML Literature on Superalignment Difficulty(2 posts)→
More from Safety
- Mantic forecasts open-weight models 9 months behind frontier in offensive cyber capability — tshevl · 2026-10-28
- Fangcun AI publishes third-party safety evaluation of DeepSeek V4.1 Flash-1017 for manipulation risks — pstAsiatech · 2026-10-28
- Study finds no real-world multilingual benchmarks for clinical AI scribes in India — DrDatta_AIIMS · 2026-10-28
- Grim Test finds AI shows no mercy for humans even while causing suffering — hollee-o · 2026-10-28
- Critic slams NYT for platforming 'TESCREAL' AI-safety circles — mjdramstead · 2026-10-28
- METR researcher joining Anthropic sparks concerns over eval lab independence — ShakeelHashim · 2026-10-28