Defining the roles of large language models (LLM) agents in the model of design

Abstract

The subject of the study is the capabilities and limitations that large language models (LLM) demonstrate when they are implemented in intellectual, technical and creative processes, in particular in design. The goal of the work is to determine the place and roles of LLM-agents in design and to develop an appropriate model of the design process augmented by LLM-agents. The article sets the following tasks: (1) to conduct a review of modern publications on approaches and methods for assessing the capabilities of large language models, in particular, when performing creative, technical and design tasks; (2) to conduct an analysis of modern approaches and methods of interaction with LLM; (3) to develop a model that defines the place and roles of LLM agents in design in interaction with the design team, the external environment and design artifacts. The following methods were used during the study: comparative-historical and retrospective analysis of the content of technical, economic, philosophical, linguistic scientific and methodological research to form a holistic vision of the current state of development of large language models and approaches to interaction with them; structural-logical analysis for the formation of a model of the design process, augmented by LLM agents. The following results were achieved: nine main groups of abilities of large language models were identified; the main modern patterns of interaction with large language models were identified; a model of the design was developed as an iterative process of knowledge enrichment, which allows describing the interaction of people and LLM-agents with each other, with the external environment and with design artifacts. The conclusions emphasize the novelty of such abilities of LLM in the family of artificial intelligence technologies as understanding and generating text in natural and programming languages, multilingualism, possession of general and industry knowledge, the ability to reason, and agency. Additionally, the need for conceptual understanding of the place and role of large language models in creative processes is highlighted. A structural model has been developed that represents design as an iterative process of knowledge enrichment and allows to define the roles of LLM-agents in interaction with the design team, artifacts, and the external environment.