OpenAI Researchers Have Discovered Why Language Models Hallucinate

OpenAI's latest research paper, "Why Language Models Hallucinate," identifies the root cause of AI hallucinations as a fundamental mismatch between training objectives and practical use: current training rewards guessing correct answers rather than acknowledging uncertainty, leading models to fabricate information when unsure. The paper suggests that revising training and evaluation methods to prioritize uncertainty acknowledgment over blind guessing could significantly reduce hallucinations, marking a critical step toward making AI chatbots reliable enough for serious economic and workflow integration.