safety verification

3 papers

• Let a Neural Network be Your Invariant
• Position: Require Frontier AI Labs To Release Small "Analog" Models
• Safely Learning Controlled Stochastic Dynamics