2 ms·Double Blind Evals: Resolving the Dual Confidentiality Dilemma in AI Safety...2 points by williamtrask 2mo ago