x
Awareness Jailbreaking: Revealing True Alignment in Evaluation-Aware Models — LessWrong