The researchers are using a technique named adversarial training to stop ChatGPT from permitting users trick it into behaving poorly (called jailbreaking). This get the job done pits various chatbots towards one another: one chatbot performs the adversary and attack