Researchers Simulated a Delusional User to Test Chatbot Safety
… They sought to find out which of the biggest LLMs are safest, and which are the most risky for encouraging delusional beliefs, in a new study published as a pre-print on the arXiv repository on April 15. …