GNoME and new materials

DeepMind predicted the structures of 2.2 million new crystals, 380 thousand of them judged stable, roughly eight hundred years of ordinary work.

The model trained on known structures and proposed new ones, scoring formation energy. DeepMind's post says external researchers had independently created 736 of the new structures in concurrent work. Critics (Cheetham and Seshadri, reported by The Register in April 2024) wrote that many of the predicted compounds do not appear very novel and that there is scant evidence for compounds combining novelty, credibility and utility; a DeepMind spokesperson answered that hundreds of the predicted materials had already been independently synthesised. The argument about the real contribution continued. Either way the prediction database became a public resource for materials science.