AI researchers trick chatbots into sharing how to make cocaine as long as they believe a user is wearing a green shirt — ‘CoT Forgery’ exploit spurs LLMs to divulge forbidden info by faking trusted chains of thought

AI researchers trick chatbots into sharing how to make cocaine as long as they believe a user is wearing a green shirt — ‘CoT Forgery’ exploit spurs LLMs to divulge forbidden info by faking trusted chains of thought

Direct: https://ift.tt/i8B6RhM

Reddit: https://ift.tt/MTpdhWP

Scroll to Top