x
Hashmarks: Privacy-Preserving Benchmarks for High-Stakes AI Evaluation — LessWrong