Openai's New O3, O4-Mini Ai Models Show Higher 'Hallucination' Rates

Madrid - OpenAI's latest AI models, O3 and O4-mini, exhibit a higher rate of 'hallucinations' compared to their predecessors. Internal tests using the PersonQA evaluation revealed that these models produce incorrect or fabricated information more frequently. The O3 model hallucinated in 33% of responses, nearly double the rate of the O1 model, while the O4-mini model reached a 48% hallucination rate. These new models are designed for tasks like programming, web navigation, and autonomous image generation. Despite their advanced capabilities, OpenAI acknowledges the issue and is actively researching the cause of the increased hallucination rates. Addressing these inaccuracies is a continuous area of focus for OpenAI, as confirmed by spokesperson Niko Felix.