Fascinating. AI “hallucinations” should more accurately be called apophenia. In AI LLM’s these “wrong” answers are actually deliberate creativity and we are leaning how to to make them less likely —for example by tuning the models to understand sarcasm so they won’t suggest using glue to stop cheese sliding off pizza.

Highly creative humans match patterns that others miss. Poorly trained and poorly wired humans match patterns incorrectly — the equivalent of using glue on pizza.