Digging Into the Internal: Causality-Based Analysis of LLM Function Calling

2025-09-18

AI Summary

Existing studies lack a deep mechanistic understanding of how function calling (FC) enhances large language models’ (LLMs’) instruction following and safety. Method: This paper introduces, for the first time, causal intervention techniques at both layer-level and token-level to systematically dissect FC’s impact on internal representations and reasoning pathways. Experiments are conducted across mainstream LLMs and two benchmark datasets. Contribution/Results: We find that FC significantly strengthens neural activations associated with compliance in critical layers, thereby improving accuracy in intent interpretation and malicious input detection. Specifically, FC achieves an average 135% performance gain over conventional prompting methods on malicious input detection—substantially boosting LLM safety robustness. Our work establishes a novel paradigm for interpretable and controllable LLM behavior regulation through fine-grained, causally grounded intervention.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Reasoning under Uncertainty: Causality

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Function calling (FC) has emerged as a powerful technique for facilitating large language models (LLMs) to interact with external systems and perform structured tasks. However, the mechanisms through which it influences model behavior remain largely under-explored. Besides, we discover that in addition to the regular usage of FC, this technique can substantially enhance the compliance of LLMs with user instructions. These observations motivate us to leverage causality, a canonical analysis method, to investigate how FC works within LLMs. In particular, we conduct layer-level and token-level causal interventions to dissect FC's impact on the model's internal computational logic when responding to user queries. Our analysis confirms the substantial influence of FC and reveals several in-depth insights into its mechanisms. To further validate our findings, we conduct extensive experiments comparing the effectiveness of FC-based instructions against conventional prompting methods. We focus on enhancing LLM safety robustness, a critical LLM application scenario, and evaluate four mainstream LLMs across two benchmark datasets. The results are striking: FC shows an average performance improvement of around 135% over conventional prompting methods in detecting malicious inputs, demonstrating its promising potential to enhance LLM reliability and capability in practical applications.

Problem

• Analyzing how function calling influences LLM behavior mechanisms
• Investigating FC's impact on model internal computational logic
• Enhancing LLM safety robustness against malicious inputs detection

Innovation

• Causality-based analysis of LLM function calling
• Layer-level and token-level causal interventions
• Function calling enhances LLM safety and robustness