Pre-Publication V3. A colloquial term for when an AI model outputs false or fabricated content that is presented as accurate.
Deliberation Summary:
(1) Consideration on how broad the term should be given it is a colloquial term. (2) Discussion on whether the terminology accurately describes how large language models behave, and if “false” or “fabricated” were being used precisely.
Pre-Publication V2. When a large language model perceives patterns or objects that are nonexistent, generating false, fabricated, misleading or inaccurate results as if they were a factual reality.
A colloquial term for when an AI model outputs false or fabricated content that is presented as accurate.
Deliberation Summary:
(1) Is language too limited as generative systems produce the same effect and generally fill in detail rather than it being a defect; is factual accuracy the right benchmark at all, or only where a system is being used to retrieve or state facts? (2) Consider objective wording for better clarity.
Pre-Publication V1. When a large language model perceives patterns or objects that are nonexistent, generating false, fabricated, misleading or inaccurate results as if they were a factual reality.