[ben_w]

No, hallucinations are specifically when they make mistakes, even when they're more like interpolation than extrapolation.

IIRC, it became a term of art during image classification AI development, where an AI might confidently assert a car was a potato, and the name stuck.

[dartos]

Idk maybe splitting hairs a bit, but I’d call any output that isn’t represented or which goes against data in a training set is a hallucination.

But there can be useful hallucinations.

Not sure about the origin of the word, but I always thought it was marketing when chatgpt and dalle was brand new

[ummonk]

I'd only call something a hallucination if the AI claims the existence of data that doesn't actually exist.

Simply making an informed guess and extrapolating to data outside the training set (whether that informed guess is is correct or incorrect) is not hallucination.