Evaluating LLMs as Interpretable Controllers for Dynamical Systems
Summary
This paper evaluates whether large language models can function as interpretable controllers for dynamical systems, specifically a thermal environment. It finds that high-complexity models like Qwen-3 14B and GPT-4o achieve accurate control and coherent reasoning, while smaller models struggle, highlighting opportunities for hybrid model-based and language-driven control strategies.
View Cached Full Text
Cached at: 07/28/26, 06:26 AM
# Evaluating LLMs as Interpretable Controllers for Dynamical Systems Source: [https://arxiv.org/abs/2607.22609](https://arxiv.org/abs/2607.22609) [View PDF](https://arxiv.org/pdf/2607.22609) > Abstract:Large Language Models \(LLMs\) are increasingly used for decision\-making and reasoning tasks, yet their potential as controllers for physical systems remains largely unexplored\. This work investigates whether LLMs can function as interpretable controllers for a dynamic thermal environment, examining their ability to follow setpoints, interpret natural\-language commands, reason about actuator effects, and incorporate prior model\-based knowledge\. Five LLMs of varying scales are evaluated under multiple scenarios, including settings with penalties on heater or fan usage and cases where the models have access to a physics\-based prediction tool\. The results show that control performance depends on model complexity: while low\- and mid\-scale models frequently misinterpret actuator dynamics or generate inconsistent reasoning, high\-complexity models such as Qwen\-3~14B and GPT\-4o achieve accurate temperature tracking, stable actuator usage, and coherent explanations aligned with physical principles\. Incorporating a physics\-based model significantly improves control smoothness and energy efficiency by enabling anticipatory decision\-making\. A detailed reasoning taxonomy further reveals a clear progression from causal misinterpretation in smaller models to cohesive and temporally aware reasoning in larger ones\. The findings demonstrate that LLMs can act as interpretable controllers when sufficiently capable and appropriately grounded in domain knowledge, highlighting promising opportunities for hybrid model\-based and language\-driven control strategies that can provide plausible explanations\. ## Submission history From: Adil Rasheed Professor \[[view email](https://arxiv.org/show-email/8a07570b/2607.22609)\] **\[v1\]**Sun, 14 Jun 2026 16:54:25 UTC \(3,182 KB\)
Similar Articles
LLMs as Post-hoc Auditors of Physiological Plausibility in Symbolic Regression: A Clinician-Evaluated Case Study
The paper investigates using Large Language Models as post-hoc auditors to evaluate symbolic regression models for interpretability and medical plausibility, with clinician assessments showing comparative model rankings are more favorably perceived than term-level interpretations.
Interpreting and Steering LLM Agents for Social Simulations
This paper explores methods to interpret and steer large language model agents in social simulations, comparing prompt-based, SAE-based, and probe-based techniques, and finds that SAE and probe methods often outperform basic prompting for control and interpretability.
Why We Need World Models for AGI: Where LLMs Fail and How World Models May Outperform
This paper argues that large language models struggle with causal reasoning and long-horizon planning due to a mismatch between sequence prediction and reasoning over latent environment dynamics, and introduces the Latent Dynamics Inference perspective along with the Flux environment to study these limitations.
DMDIntel: Interpreting Large Language Models via Dynamic Mode Decomposition
This paper introduces DMDIntel, a method that uses dynamic mode decomposition to make LLM predictions interpretable by decomposing hidden states into modes and ranking token attributions, outperforming existing techniques like PCA, integrated gradients, and SHAP.
Understanding Large Language Models
This chapter reviews current understanding of Large Language Models, discussing their Transformer architecture, emergent capabilities resembling human cognition, and debates about whether LLMs genuinely understand or merely simulate understanding.