1 paper
Jonathn Chang, Leonhard Piff, Suvadip Sana +2
Aligning AI with human values is a pressing unsolved problem. To address the lack of quantitative metrics for value alignment, we propose EigenBench: a black-box method for compara…