LessWrong AI
· Communities
Attackers Can Subliminally Implant a Backdoor at Low Sample Count Without Prompt Access
Work done at Redwood Research, quick update on results found as part of a larger project. Thanks to @SebastianP and @egan for comments on drafts.TL;DRChanging the teacher for only 100 (0.5% of) completions in fine-tuning can allow attackers to covertly implant a backdoor without control of the dataset prompts. This dat