Transformers as a Physical Model in AI

The structure of a transformer (an artificial intelligence model based on attention models) is considered. Analogies with physical models are discussed (‘‘transformer as an evolution operator for a system of attention models as a Hamiltonian’’), with the approaches by Yu.I. Manin (‘‘renormalization and computations’’) and M. Marcolli (‘‘generative linguistics as algebraic and physical model’’), and with the concept of a genome as a ‘‘gas of interacting genes’’ (E.V. Koonin).