English Español Français Deutsch
← Back to the list
Hallucination and factuality Methodologists Directly upstream of ai100's methodology

Yue Zhang

Organizing what the field calls "hallucination" into categories that actually mean different things.

Before 2023, "LLM hallucination" was a single bucket — a wrong fact, a made-up citation, a confused chain of reasoning, all got the same word. Zhang's "Siren's Song in the AI Ocean" survey separated the categories with the discipline of a taxonomist: factuality hallucinations (the model contradicts the world), faithfulness hallucinations (the model contradicts its own context), and several finer subcategories that have since become standard vocabulary in the literature. Reading him is the fastest way to stop saying "hallucination" as if it were one phenomenon.

Worth following when
you need to talk about LLM hallucination with technical precision rather than as a single hand-wave term.
Topics
taxonomy of LLM hallucination; faithfulness versus factuality; survey methodology in NLP.
Key works
"Siren's Song in the AI Ocean: A Survey on Hallucination in LLMs" (2023); ongoing Westlake NLP group output on structured prediction and LLM evaluation.