Physicists Say Formula Can Predict When AI Chatbots Turn Bad

Physicists at George Washington University have developed a formula that can estimate when an AI chatbot will shift from good answers to bad ones. Early tests on small models support the approach.