Pieces
[last updated]
Safe by Design, by not Desiring 5 Feb, 2026
Outline of an approach towards safe(r) AI, based on the disinterested pursuit of truth.
Written with Damiano Fornasiere and others, for LawZero
Superintelligent Agents pose Catastrophic Risks: Can Scientist AI Offer a Safer Path? 25 Feb, 2025
A review of the risks of deploying highly intelligent agents, an analysis of agency, and a rough sketch of some ideas for how one might approach the problem, which Bengio has branded “Scientist AI”, and has become the motivating force behind LawZero. Co-written with many others.