Hallucination
A hallucination is a confident, fluent output that no source supports. It is usually described as a model defect, but the more useful description is a context gap: the model cannot perceive a missing piece as missing, so it experiences the gap as an ordinary generation problem and fills it. That is why better prompting does not fix it and why the fix has to sit outside the model.
Related terms
Where this term is used
On The AI Brain
Research
When Context Collapses: What a Geopolitical Crisis Revealed About AI's *Missing Layer*
During the Iran-US escalation cycle, we ran frontier AI models against the same questions under different context architectures. Five failure modes emerged that no model upgrade will fix. They are pr…
Research
The AI Audit Panel: verification built into the architecture
How COVE — our production implementation of Chain-of-Verification (CoVe, Meta AI / ETH Zurich) — reduces hallucinations ~35% and makes every claim auditable.