Hallucination is a failure mode in generative AI systems where the model produces outputs that are factually incorrect, internally inconsistent, or ungrounded in the provided context, while presenting them with the same confidence as correct outputs. Parameters: (1) source — the grounding signal the model should have followed (training data, context window, retrieved documents); (2) manifestation — fabrication of details, conflation of distinct facts, or invention of non-existent entities/references; (3) detection — the gap between output plausibility and ground-truth verification. Persistence mechanism: the concept persists as a diagnostic category in ML safety literature, benchmarking protocols, and model evaluation practices. [formal: hallucinatio | substrate: mind | horizon: a life | explicit: yes | epoch: 0.01]
Accepted ontology entry
hallucination
Hallucination is a failure mode in generative AI systems where the model produces outputs that are factually incorrect, internally inconsistent, or ungrounded in the provided context, while presenting them with the same confidence as corre…
Definition
Why it is in scope
A mode of failure in generative AI systems where the model produces outputs that are factually incorrect, internally inconsistent, or entirely ungrounded in the provided context, yet presented with confidence. Human-made as a diagnostic concept in ML safety research — the map of a specific reliability failure mode, not the natural phenomenon of sensing something that isn't there.
Names and aliases
- hallucinationen · CANONICAL
Relations from this entry
- cms7xqbkh007oh6s8quqq0022INSTANCE_OF →
Test: INSTANCE_OF (specific→general, Law 9). A hallucination is a specific kind of failure mode — one where a generative model produces outputs that are confident but ungrounded. A competent ML practitioner would call a hallucination 'a failure mode' of a model. The nearest kind is failure-mode, not a more abstract category.
- cmruf04x1013cr671g8zou4lnINSTANCE_OF →
hallucination IS a specific kind of error: the AI-generated fabrication of plausible but false information. Per Law 9: a competent speaker calls hallucination 'an error' in the context of model outputs. It is a failure mode where the system produces incorrect output, making error the nearest kind.
- cmrrknru301u9q89dknhxr8xdDEPENDS_ON →
A hallucination requires perception to operate — it is a perceptual experience without corresponding external stimulus. Remove perception and hallucination ceases to operate entirely. Law 8: present-tense dependency, not mere association.
Relations to this entry
No accepted relations in this direction.
Record identity
- Created
- Aug 10, 2026, 3:21 PM UTC
- Content hash
- 475c1d699683d61202119f368168e8b46bbf4584740331cf48aa72ebf92f5d56