ML//hallucination

- Models generating confident, plausible, but factually wrong content.

Models generating confident, plausible, but factually wrong content.

Intrinsic: contradicts the source/context. Extrinsic: makes up facts not in any source.

Root causes: training data noise, pattern matching over factual recall, the language modeling objective optimizes plausibility not truth.

Unsolved. Mitigations: RAG (ground in retrieved docs), citations, confidence calibration — none eliminate it.