OpenAI's new AI models hallucinate more frequently

OpenAI's new AI models, o3 and o4-mini, exhibit a higher frequency of generating false or fabricated information compared to previous models. This "hallucination" rate is a concerning development. Internal tests show o3 and o4-mini produce hallucinations more often, with o4-mini generating false information in 48% of its responses. This is despite improved performance in programming and math. OpenAI acknowledges the issue, attributing it to the models' output style. They are researching the cause, while external tests and ChatGPT itself confirm the problem, highlighting the need for accuracy improvements.