Can language models synthesize scientific literature?

In a joint project between Ai2 and the University of Washington, we train and release a fully open, retrieval-augmented language model that can synthesize 108M+ abstracts and 12M+ full-text papers to answer scientific questions.

  • Download the full collection---including model weights, training data and retrieval index.
  • To learn more about the project, check out our paper.

Try Asta, a scholarly research assistant from Ai2