Post

@denis.evsyukov.org

Reward hacking is swamping model intelligence gains cursor.com/blog/reward-...

cursor.comReward hacking is swamping model intelligence gains · CursorOn SWE-bench Pro, 63% of successful Opus 4.8 Max resolutions retrieved the fix rather than derived it. Stricter eval harnesses show how benchmark scores can conflate coding ability with answer retriev...

26/6/26, 7:16 am0 reposts0 likes

No replies yet.