SMALL MOLECULE OPTIMIZATION WITH LARGE LANGUAGE MODELS

Summary from the TLDR AI newsletter

Two new language models were able to achieve an 8% improvement on the Practical Molecular Optimization benchmark by collecting a dataset of 100m molecules (40B token equivalent).