Do Large Language Models Dream of Sockets?
Open Access
- 20 July 2024
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 103-105
Abstract
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. We present our preliminary results, including how LLMs understand the behavior of example systems such as web servers. Our contribution focuses on testing how and how well an LLM can diagnose protocol traces, receive and send protocol messages over sockets, and handle complex protocol fields and interfaces.Keywords
This publication has 11 references indexed in Scilit:
- Large Language Model guided Protocol Fuzzing2024
- PROSPER2023
- Generative, High-Fidelity Network Traces2023
- Rethinking data-driven networking with foundation models2022
- Practical GAN-based synthetic IP header trace generation using NetShare2022
- Semi-automated protocol disambiguation and code generation2021
- Using Network Traces to Generate Models for Automatic Network Application Protocols Diagnostics2019
- A Survey on Function and System Call Hooking ApproachesJournal of Hardware and Systems Security, 2017
- Extension Mechanisms for DNS (EDNS(0))2013
- Inferring Protocol State Machine from Network Traces: A Probabilistic Approach2011