Galactica AI: Meta's Scientific Language Model Explained

Galactica is a large language model developed by Meta AI and Papers with Code, trained specifically on a vast corpus of humanity's scientific knowledge. Unlike general-purpose chatbots, Galactica was engineered to serve as a specialized interface for navigating dense scientific literature, papers, and data—effectively acting as a research companion capable of summarizing and manipulating complex information about the natural world. The project was launched as an open research initiative aimed at gathering community feedback on how large language models perform in scientific domains. However, the public-facing demo was pulled shortly after release once it became clear that the model could generate confident-sounding but factually incorrect scientific text, raising concerns about misuse beyond its intended research audience. While the interactive demo is no longer accessible to the general public, the underlying models and research paper remain available to academics and researchers who wish to study the architecture, reproduce published results, or build upon the work for further scientific NLP research.

The public demo of Galactica is no longer available for general use. Researchers interested in the project can access the published models and accompanying paper to study the methodology, replicate the reported results, or adapt the underlying techniques for their own scientific NLP experiments.