AI 'Hallucination' Is Actually Three Distinct Problems, Not One

The term 'hallucination' is widely used to describe AI errors, but a software developer's hands-on experience suggests it conflates at least three fundamentally different failure types. The first involves missing context that exists and can be supplied, making the error correctable through better prompting or added information. The second concerns unverifiable claims about a model's internal states, such as whether it truly 'understands' emotions, where no instrument currently exists to confirm or deny the output. The third involves predictions about future events where the required information does not yet exist for anyone, making no prompt or technique capable of fixing the gap. Research from OpenAI and Georgia Tech has similarly examined why language models produce false outputs, reinforcing the case that treating all hallucinations as a single bug leads to solutions that fail to converge.