project-014
magnets need to move into the 21st century. the NdFeB incumbent, which dominates over 60% of a $30B magnet market, was discovered in the 80s… surely we can do better.
project-14 aims to create novel generative and machine learning methods for discovering the next generation of magnetic materials.
curie temperature, thermodynamic stability, magneto-crystalline anisotropy energy, and total magnetic density are already being accurately predicted by project-14’s models.
an adapted monte carlo tree search algorithm, a custom scoring function built from the project-14 property predictors, and an ensemble of inorganic crystal-specific generative models define a reinforcement learning environment that empowers the leading large language models to navigate the complex space of potential new magnets.
more to come…