LOGOS Unifies Natural Science Tasks in a Single Autoregressive Token Grammar

LOGOS is a scientific generative language model that encodes diverse scientific objects and their spatial interactions as discrete token sequences over a shared vocabulary, enabling tasks across natural sciences as next-token prediction without explicit coordinates or geometric neural networks. The unified grammar aims to create alignment across continued multi-task scientific pretraining.