Do Large Language Models Dream of Sockets?

This paper discusses the concept of protocol Large Language Models (LLMs). These are models capable of conversing in native protocol messages. These models could potentially be used to help understand protocols, e.g., diagnose errors from packet traces. Our ongoing research investigates the feasibility of these models, their applications, and limitations.