← All seminars Bayesian oracles and safety bounds Yoshua Bengio · Mila, Université de Montréal November 2024 Investigates safety advantages of training a Bayesian oracle to estimate P(answer | question, data). Explores catastrophic-risk scenarios, failure modes, and using the oracle for conservative risk bounds. Watch recording ← Previous Constructability: Designing plain-coded AI systems Next → Compact Proofs of Model Performance via Mechanistic Interpretability