English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

A Framework for Context Enrichment and Selective Robustness in LLM Social Simulations

Forum topic · 小凯 · 2026-11-05

Summary

Researchers use large language models as agents in social simulations, but LLM agents have been shown to change behavior substantially under meaning-preserving prompt variations, creating a validation problem. A plausible hypothesis attributes this brittleness to a lack of grounding that could be reduced by enriching agent context. This paper addresses the question with two contributions: controlled context enrichment, a set of prompt substitution rules that systematically enrich distinct types of context (neutral fillers, generic social background, descriptive decision grounding, and normative decision grounding); and selective robustness, a notion requiring insensitivity to meaning-preserving prompt variations while producing directional responses to decision-relevant changes in the social world. The framework is demonstrated in a prisoner's dilemma-like game, showing how to identify which enrichment forms yield better and more robust agent behavior. Different enrichment forms produced distinct behavioral patterns, differing across the two tested models. Source: arXiv 2611.06914, by Mattia Greiche, Reihaneh Rabbany, and Maximilian Puelma Touzel.

Paper Overview

Field: NLP Authors: Mattia Greiche, Reihaneh Rabbany, Maximilian Puelma Touzel Published: 2026-11-04 arXiv: 2611.06914

Summary

Researchers use large language models as agents in social simulations, where agents are prompted with observations and background information intended to shape their actions. Yet LLM agents have been shown to substantially change their behavior under meaning-preserving prompt variations, creating a validation problem for social simulation.

A plausible hypothesis is that this brittleness reflects a lack of grounding and can be reduced by enriching the context provided to the agents. However, whether richer context improves robustness remains unclear, in part because context richness and robustness are rarely defined in ways that support controlled comparisons.

Key Contributions

1. Controlled context enrichment: a set of prompt substitution rules that systematically enrich distinct types of context:

  • neutral fillers
  • generic social background
  • descriptive decision grounding
  • normative decision grounding
2. Selective robustness: a notion of robustness requiring insensitivity to meaning-preserving prompt variations, while still producing directional responses to decision-relevant changes in the social world.

Findings

The framework is demonstrated in a prisoner's dilemma-like game, showing how to identify which forms of context enrichment produce better and more robust agent behavior. Different forms of enrichment led to distinct behavioral patterns, and these patterns differed between the two tested models.

--- *Auto-collected on 2026-11-06*

Tags

#llm-agents#social-simulation#robustness#prompting#nlp#arxiv#validation

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178636787