Your model already knows the answer: how benchmark answers leak into LLMs from Hacker News on 2026-08-05 16:56 (#77FYF) Comments
