Meta, OpenAI, Anthropic and Cohere A.I. models all make stuff up — here's which is worst
OpenAI's GPT-4 performs well in terms of math and hallucination rates compared to other AI models, but it is less effective at hedging its answers, according to a report by Arthur AI.