LLMs Unexpectedly Understand Molecular Shapes and Stability

New research shows that some advanced large language models (LLMs) have unexpectedly developed an ability to understand and reason about the 3D shapes and stability of molecules, performing better than some traditional chemistry tools.