Аннотация: (English) The theoretical and practical foundations of designing physical data access plans using large language models are explored. The structure of a physical plan, main data access techniques, and connection algorithms are explained. The potential of language models for transforming SQL queries, selecting control instructions, and generating execution options is revealed. A sequence for creating a physical plan is provided, which involves analyzing the original query, selecting a transformation rule, generating a new query, verifying its syntax and semantics, creating a plan with a database management system, and evaluating performance. It is determined that using language models necessitates mandatory verification of query equivalence, operation validity, and actual execution time. Conditions under which the expense of using the model may be offset by increased data processing efficiency are identified.