build1 publisher
A 4B model learned two-hop geography with its top J-lens directions suppressed all through training
In a single-author LessWrong experiment, accuracy on held-out countries went from 20.9% to 97.7% with the lesion active in every forward pass, and the recovery survived refitting the lens to the adapted model.
Publishers:lesswrong.com
Reality
- Evidence34
- Adoption12
- Hype gap+22
- Incentives18
- Confidence42