On July 3, Alibaba DAMO Academy announced that its AI intelligent agent ElementsClaw, developed in conjunction with Renmin University of China, University of Chinese Academy of Sciences and other institutions, successfully independently discovered and experimentally verified four new superconducting materials. Relevant results are published on the arXiv preprint platform.
According to DAMO Academy, ElementsClaw adopts a "specialized fusion" architecture: a 1B parameter atomic basic model is pre-trained based on 125 million molecular and crystal structure data, and the predicted superconductivity AUC reaches 0.996, and the average error of the critical temperature is within 1K.AI used only 28 GPU hours to complete the screening of 2.4 million crystal structures and predict 68,000 superconducting candidate materials.The four materials verified by the team include Hf21Re25, Zr4VRe7, HfZrRe4 and Zr3ScRe8, with critical temperatures up to 6.5K.
Rong Yu, head of scientific intelligence at DAMO Academy, said that this is the first batch of superconducting materials discovered and verified by AI agents, verifying the potential of this framework in the field of material discovery. Huang Wenbing, associate professor at the Hillhouse School of Artificial Intelligence at Renmin University of China, said that the AI agent is also expected to be used in the discovery of new materials such as solid-state battery electrolytes and heterogeneous catalysts. At present, DAMO Academy has opened the relevant prediction database for free use by scientific researchers.