Yassine Essifi

This is a crosspost from the General-Purpose AI Policy Lab research blog. The Epoch Capabilities Index compresses many benchmark scores into one for each model, following the framework of the Rosetta Stone paper. In a previous post, we added human baselines to the same scale to see how the models...

TL;DR: Evaluation awareness, when a model recognizes it is being tested rather than deployed, is being reported in frontier models with increasing frequency and has started appearing in external safety reviews such as Apollo's and METR's reviews of Opus 4.6. It is also becoming harder to detect, with the most...