Hallucination
A fluent, confident model output that is simply wrong, invented rather than recalled.
Because a language model predicts what reads as plausible rather than checking a store of facts, it can state invented citations, dates, or events in perfect prose. The term is borrowed from psychology and is a little misleading: the model is not perceiving anything wrongly, it is generating text that sounds right. Entries in this record note hallucinations where they changed how people or institutions treated the technology.
Where it mattered in the record
-
Meta withdraws its Galactica science model after three days
Meta AI released Galactica on 15 November 2022, a language model trained on scientific text and intended to help write papers and summaries, with a public demo. Users showed it producing fluent but false claims and invented citations, and Meta took the demo down on 17 November.