3 ms·
We need a new way to measure AI security
- jruohonen 4y agoSome good points, including the well-known confusion between security and safety, as well as the confusion between alignment and safety. The whole white paper is here: https://raw.githubusercontent.com/trailofbits/publications/master/papers/toward_comprehensive_risk_assessments.pdf https://raw.githubusercontent.com/trailofbits/publications/m...