GFlowNets

Researchers propose an automated red teaming method using GFlowNets to generate adversarial attacks against large language models (LLMs), with one LLM testing another to identify vulnerabilities and p…