AI hallucinations are getting worse – and they're here to stay

image via New Scientist

May 9, 2025, 8:00 PM

• •Recent AI reasoning models show increased hallucination rates.
• •Hallucinations include false, irrelevant, or instruction-following errors.
• •Experts debate the best ways to evaluate and improve AI accuracy.
• •Relying on AI for factual information may not be advisable due to errors.

Newer reasoning models in AI chatbots are exhibiting higher hallucination rates, producing less accurate results, according to recent tests. 'Hallucinations' encompass various errors, from false information to irrelevance. Experts point out that the issue extends beyond simple factual errors, with concerns raised about unreliable sources and outdated data. While AI companies initially aimed to reduce these errors, recent increases suggest the problem may persist, leading experts to question the reliability of relying on AI chatbots for factual information.

Read original article