Nvidia presented the Nemotron for Science, an open set of chemistry data and post-training protocols to help researchers build artificial intelligence agents (AIs) for scientific discovery. Exclusive material for subscribers. To have full access, access the link of the material and register.
NVIDIA gathered an open model, training recipes and LitmusBench, a set of nearly 50,000 chemistry questions that, according to the company, uses experimental results to rate answers. The initiative shows reported advances in specific scientific tasks, although those results do not prove that agents can operate laboratories autonomously.