AI is Hallucinating Increasingly

OpenAI's latest AI models, o3 and o4-mini, launched in April 2025, excel in complex tasks like math and coding but hallucinate more than their predecessors, with o3 at a 33% hallucination rate and o4-mini at 48% on the PersonQA benchmark, compared to 16% and 14.8% for older models o1 and o3-mini. This increase, reported by TechCrunch, reverses the trend of decreasing hallucinations in newer models. OpenAI admits it doesn't fully understand why this is happening, noting that more research is needed. The models' tendency to generate false information, such as broken links or fabricated actions, undermines their reliability, especially for accuracy-critical applications. OpenAI is working to address this ongoing industry-wide issue.

https://futurism.com/the-byte/openai-new-ai-problem-hallucinate-more