Teaching LLMs to Be Deceptive
Schneier on Security
FEBRUARY 7, 2024
Interesting research: “ Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training “: Abstract: Humans are capable of strategically deceptive behavior: behaving helpfully in most situations, but then behaving very differently in order to pursue alternative objectives when given the opportunity. If an AI system learned such a deceptive strategy, could we detect it and remove it using current state-of-the-art safety training techniques?
Let's personalize your content