Controlling Tool Use with Heading-Specific Activation Steering
This work provides insights into the internal representations of tool-use in LLMs, but the findings are incremental and raise more questions than answers.
The paper investigates whether tool-use decisions in tool-augmented LLMs have stable internal representations that can be manipulated via activation steering, finding that steering vectors from heading-anchor positions can suppress unnecessary tool use, but geometric analysis reveals diffuse, bimodal alignment rather than clean linear structure, indicating that tool-use representations are geometrically irregular.
Tool-augmented large language models extend their capabilities beyond parametric knowledge through external tools, but tend to invoke them unnecessarily. We investigate whether tool-use decisions have any stable internal representation that can be extracted and manipulated, a question that is non-trivial given that tools exist entirely in context at inference time and have no direct encoding in model weights. We show that steering vectors extracted from heading-anchors positions exert bidirectional causal control over tool-invocation behavior across five open-source models and three domains, suppressing unnecessary tool use most effectively in domains where parametric reasoning suffices. However, geometric analysis reveals that this causal effectiveness does not correspond to clean linear structure: tool-invocation steps exhibit diffuse, bimodal alignment with the suppression vector rather than the consistent negative alignment a linear encoding account would predict, and different tool types recruit largely distinct internal signatures with low cross-tool feature overlap. We hypothesize these geometric properties are indicative of the non-parametric nature of tools, and distinguish tool-use steering vectors from those extracted for parametrically grounded concepts. The relationship between this geometric irregularity and the observed causal effectiveness remains an open question.