Meta AI released Galactica on 15 November 2022 as a system intended to store, combine, and reason about scientific knowledge. Meta warned that its outputs could be unreliable because language models hallucinate, but users quickly demonstrated that the model could generate authoritative-looking falsehoods. In one example, Galactica created a paper about avatar production and cited a nonexistent study attributed to a real researcher working in the field.
Users also reported offensive and inaccurate material, exposing the gap between the product’s scientific presentation and its reliability in practice. Meta withdrew Galactica on 17 November, only two days after release. The rapid reversal became an early example of a major technology company publicly launching a generative model despite known fabrication risks, then relying on a warning label that proved insufficient to prevent misleading outputs or reputational damage.
