Your model already knows the answer: how benchmark answers leak into LLMs 12 points by fran-mora 5 hours ago story
Hacker News