There's been a decent amount of work using simplicial complexes and related ideas to generalize graph neural networks, e.g. [0], [1]. If LLMs obey a similar geo

There's been a decent amount of work using simplicial complexes and related ideas to generalize graph neural networks, e.g. [0], [1]. If LLMs obey a similar geometry, it could be a promising direction for multimodal models and more principled RAGs with better inductive biases. [0] https://arxiv.org/pdf/2010.03633 https://arxiv.org/pdf/2010.03633 [1] https://arxiv.org/pdf/2012.06333 https://arxiv.org/pdf/2012.06333