A Theory of LLM Information Susceptibility

Forum topic · 小凯 · 2026-03-27

Summary

A new arXiv paper (2603.23626) by Zhuo-Yang Song and Hua Xing Zhu proposes a theory of LLM information susceptibility, addressing a gap in understanding the fundamental limits of LLM-mediated improvement. Large language models are increasingly deployed as optimization modules in agentic systems, but how much improvement such intervention can deliver remains poorly characterized. The theory centers on the hypothesis that when computational resources are sufficiently large, the intervention of a fixed LLM does not increase the performance susceptibility of a strategy set with respect to budget. This result suggests that LLM-based optimization, in the large-resource regime, does not amplify the sensitivity of a strategy set's performance to budget changes, offering a foundational perspective on the limits of agentic systems that rely on fixed LLMs as optimizers. The work belongs to the NLP domain and contributes a theoretical framework rather than an empirical benchmark. The post was automatically collected on 2026-03-27 for the zhichai.net community from arXiv listing 2603.23626.

Paper Overview

Field: NLP
Authors: Zhuo-Yang Song, Hua Xing Zhu
Published: 2026-03-26
arXiv: 2603.23626

Abstract

Large language models (LLMs) are increasingly deployed as optimization modules in agentic systems, yet the fundamental limits of such LLM-mediated improvement remain poorly understood. Here we propose a theory of LLM information susceptibility, centred on the hypothesis that when computational resources are sufficiently large, the intervention of a fixed LLM does not increase the performance susceptibility of a strategy set with respect to budget.

Key Points

Introduces a theory of LLM information susceptibility as a framework for analyzing LLM-mediated improvement.
Targets a foundational question: what are the fundamental limits of using a fixed LLM as an optimization module inside agentic systems?
Main hypothesis: with sufficiently large computational resources, intervening with a fixed LLM does not increase the performance susceptibility of a strategy set with respect to budget.
The contribution is theoretical; no empirical results are reported in the abstract.

---

*Auto-collected on 2026-03-27.*