Meta, OpenAI, Anthropic and Cohere A.I. models all make stuff up — here’s which is worst

Researchers from Arthur AI tested top AI models and found that some make up facts, or “hallucinate,” significantly more than others.