“Hallucinations” are getting worse in “reasoning” models from OpenAI, Google, and DeepSeek. Experts struggle to explain why. (Cade Metz, Karen Weise, New York Times)
Summary
- A.I. systems' "hallucinations" are getting worse despite companies' best efforts..
- New "reasoning" models from OpenAI, Google, and DeepSeek hallucinate at rates up to 79%.
- Experts struggle to explain why.
