liveBook · Manning

Intrinsic hallucination

As explained in Introduction to Generative AI, Second Edition

An intrinsic hallucination is a model-generated response that is inconsistent with information in its source material.

Suppose source material says an event happened in 2006, but a language model says it happened in 2016. That contradiction is an intrinsic hallucination: the response is inconsistent with information in its source material. section 5.7

This differs from an extrinsic hallucination, which cannot be confirmed in the source material. A model may guess when information is limited or conflicting, so a confident-sounding answer still needs checking against its source. section 5.7

How Introduction to Generative AI, Second Edition covers it

How deep it goes
Defines the inconsistency and offers uncertain guessing as an explanation, without giving a detection procedure.
Examples
Uses a hypothetical source date of 2006 and a model answer of 2016 for the same event.
What you can do after
Evaluate/understand it by checking whether a model’s claim conflicts with available source material.
How it connects
Connects intrinsic hallucination to extrinsic hallucination, source material, and uncertainty in large language models.
Cover of Introduction to Generative AI, Second Edition
liveBookmanning.com
Read Introduction to Generative AI, Second Edition on liveBook Get the book at manning.com

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Introduction to Generative AI, Second Edition. The section links point to the parts of the book it draws on.