Story perspectives
OpenAI's New Models Face Rising Hallucination Rates
4/19/2025
46 10
1 of 1
Story summary
- OpenAI's o3 and o4-mini AI models show higher hallucination rates, with o3 at 33% and o4-mini at 48% on PersonQA.
- While coding and math performance has improved, accuracy suffers due to frequent inaccuracies.
- OpenAI recognizes the necessity for further research on increasing hallucinations.
- The industry is pivoting towards reasoning models, which may enhance performance but risk more hallucinations.
- Implementing web search capabilities could enhance accuracy and mitigate hallucinations.
