[Paper Review] A Guide to Large Language Models in Modeling and Simulation: From Core Techniques to Critical Challenges

Philippe J. Giabbanelli|arXiv (Cornell University)|Feb 5, 2026

Artificial Intelligence in Healthcare and Education0 citations

TL;DR

This paper provides a practitioner-focused guide on using LLMs in Modeling & Simulation, detailing core techniques, common pitfalls, and practical considerations across prompting, hyper-parameters, augmentation, and evaluation.

ABSTRACT

Large language models (LLMs) have rapidly become familiar tools to researchers and practitioners. Concepts such as prompting, temperature, or few-shot examples are now widely recognized, and LLMs are increasingly used in Modeling & Simulation (M&S) workflows. However, practices that appear straightforward may introduce subtle issues, unnecessary complexity, or may even lead to inferior results. Adding more data can backfire (e.g., deteriorating performance through model collapse or inadvertently wiping out existing guardrails), spending time on fine-tuning a model can be unnecessary without a prior assessment of what it already knows, setting the temperature to 0 is not sufficient to make LLMs deterministic, providing a large volume of M&S data as input can be excessive (LLMs cannot attend to everything) but naive simplifications can lose information. We aim to provide comprehensive and practical guidance on how to use LLMs, with an emphasis on M&S applications. We discuss common sources of confusion, including non-determinism, knowledge augmentation (including RAG and LoRA), decomposition of M&S data, and hyper-parameter settings. We emphasize principled design choices, diagnostic strategies, and empirical evaluation, with the goal of helping modelers make informed decisions about when, how, and whether to rely on LLMs.

Motivation & Objective

• Clarify how LLMs fit into modeling and simulation pipelines and what practitioners need to know to use them effectively.
• Highlight engineering decisions in prompting, data handling, and evaluation that affect replicability and transparency.
• Identify common misconceptions and non-determinism issues that impact M&S outcomes.
• Discuss emerging concepts like forgetting, role prompting, and multimodal inputs in LLM-based M&S systems.
• Offer practical exercises and guidelines to decompose problems and assess LLM knowledge in M&S contexts.

Proposed method

• Review fundamental LLM components such as prompting and hyper-parameters.
• Explain augmentation techniques like Retrieval-Augmented Generation (RAG) and contextual knowledge.
• Discuss non-determinism sources and mitigation strategies.
• Warn against over-reliance on LLMs and suboptimal reinventions of solutions.
• Describe representation choices for models and outputs and their impact on performance.
• Provide practical guidance on prompt communication, reproducibility, and automation of prompt design.

Experimental results

Research questions

• RQ1 What are the core components and engineering decisions involved in effectively using LLMs for modeling and simulation?
• RQ2 How do prompting strategies, hyper-parameters, and augmentation techniques affect replicability, performance, and transparency in M&S tasks?
• RQ3 What common mistakes and misbeliefs surround LLMs in M&S, and how can they be mitigated?
• RQ4 How do representations of model structure and outputs influence LLM performance in M&S settings?
• RQ5 What practical guidelines and exercises can help researchers reason critically about LLM-enabled M&S workflows?

Key findings

• Prompts are central interfaces; task decomposition, validation prompts, and explicit task definitions improve reliability.
• Decoding hyper-parameters like temperature and sampling strategies significantly influence outputs and reproducibility, with effects varying by LLM and task.
• Contextual augmentation (RAG) can help or hinder performance depending on retrieval quality and integration, necessitating careful prompting and evaluation.
• Representation of models (e.g., edge lists vs adjacency matrices) can materially affect LLM performance; empirical evaluation across representations is advised.
• Non-determinism sources extend beyond temperature; thorough evaluation and mitigation are required for robust M&S results.
• Automation in prompt design is growing, but human input remains essential for task specification and interpretation.