AI models can be dangerous before public deployment - METR

Why pre-deployment testing is not an adequate framework for AI risk management

saved by

• Alex Yun
• Jackson Kozlowski