AI delusion is a user’s unjustified confidence in generative AI: treating its outputs as infallible or authoritative and responding with too little skepticism or cross-verification. It differs from AI hallucination, which describes model behavior rather than the user’s reaction. section 8.3
When the model reflects bad training data
If training data says the sky is green, a model that repeats this claim is not necessarily hallucinating; it may be reproducing its training data. Accepting the claim without checking it illustrates the user-side risk of AI delusion. section 8.3
