@repligate Are we already seeing models that are smart enough to deliberately score high on selected evaluations while concealing their true capabilities?
in reply to: 1897057790520156669
same thread: 1897057790520156669 1897060658526757295 1897061145078599728 1897061287483625673 1897062097969930272 1897089814371164446 1897091048687722932 1897097325639500254 1897097451867385981 1897097586017992997 1897278844383715622 1897279307392893089 1897282395373756889
Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.