LESSWRONG
LW

2971
ananana
0010
Message
Dialogue
Subscribe

Posts

Sorted by New

Wikitag Contributions

Comments

Sorted by
Newest
No posts to display.
No wikitag contributions to display.
Refusal in LLMs is mediated by a single direction
ananana1y10

Nice work! Can the reason these concepts are possible to 'remove' be traced back to the LoRA finetune?

Reply