Galactica

Galactica is a large language model trained exclusively on a massive corpus of scientific literature, textbooks, encyclopedias, and academic knowledge. Developed by Meta AI and released in November 2022, it was designed to store, combine, and reason about scientific knowledge — with the goal of helping scientists navigate and synthesize the exponentially growing body of academic research.

The model was trained on 48 million scientific papers, academic textbooks, scientific websites, encyclopedias, and reference materials. This specialized training made it unusually competent at scientific question answering, summarizing research papers, writing scientific content, and performing academic reasoning tasks compared to general-purpose LLMs trained on diverse web text.

Galactica was controversially taken offline just three days after its public demo in 2022 after demonstrations showed it confidently generating plausible-sounding but factually incorrect scientific text — a phenomenon that illustrated the fundamental challenge of AI hallucination in high-stakes domains. The model remains available for research use and continues to inform work on more reliable scientific AI systems.