Tweet by Jack_W_Lindsey
July 31, 2025
We can also show the mechanism behind how models answer multiple choice questions — “I need an answer” features go hunting for “correct answer” features on the appropriate answer tokens! (Though we still don’t know where those come from) https://t.co/tGSNdhMIJk
- Author
- Jack_W_Lindsey
- Date
- July 31, 2025
- Canonical URL
- /tweets/jack-w-lindsey-1950952357191352656-73c4c1