Top interpretability expert Neel Nanda: don't rely on us to save you on the current trajectory
burny_tech · x · 2026-09-30
PalisadeAI amplified a stark warning from Neel Nanda, widely regarded as one of the world's top two mechanistic interpretability experts: "Speaking as an interpretability expert, please do not rely on us to save you on the current trajectory." A rare pessimistic signal from a leading alignment researcher about whether current interpretability tools can keep up.
More from AGI Musings
- alphaXiv builds agent preprint server to study autonomous research agents — SRSchmidgall · 2026-09-30
- Ex-DeepMind AGI deployment lead: 1,386 frontier lab employees signed safety letter, public discourse understates internal concern — Miles_Brundage · 2026-09-30
- Why Millions Are Turning to AI for Moral and Spiritual Advice — adariostrange · 2026-09-30
- 63% of Americans See Real AI Risk, So "Doomer" Label Is Overused — DKokotajlo · 2026-09-30
- "Inference is computation, not thinking": the LLM reasoning debate flares again — gerardsans · 2026-09-30
- Dev argues models can't extrapolate user preferences even with unlimited perfect context — gabriel1 · 2026-09-30