Researchers Use AI to Jailbreak ChatGPT, Other LLMs

“Tree of Attacks With Pruning” is the latest in a growing string of methods for eliciting unintended behavior from a large language model.