Framework reveals challenges in fixing large language model decisions
When Does Correction Become Repair? Mechanistic Auditing of Internal Interventions in Tool-Using LLMs
Computation and LanguageSoftware Engineering
Summary
Large language models (LLMs) often decide whether to use an external tool, answer directly, or ask for clarification. The authors found that trying to fix LLMs' internal decision-making can sometimes cause new mistakes instead of true improvement. They created a tool called SAKIKO to carefully check if these fixes actually improve LLM behavior or just change it in confusing ways. This shows that it’s important to verify the final outcomes before claiming an internal fix works.
What this means in practice
- •For ai system engineers: Use mechanistic auditing to evaluate if internal intervention methods truly improve decision-making in tool-using language models.
- •For machine learning developers: Improve development workflows by detecting collateral damage when altering internal model activations before tool execution.
Authors
Jiayi Li, Ruizhe Li
Abstract
Before invoking external tools, an agentic LLM must select among a K-way action space: executing a call, seeking clarification, answering directly, or declining. While internal activation steering can alter these pre-execution decisions, conventional aggregate metrics obscure where altered states land and what collateral damage they inflict. We present SAKIKO, an auditing framework that formalizes representation repair via directional error discovery, router-conditioned intervention, destination-resolved verification, and prospectively frozen statistical licensing. Across seven LLMs on When2Call and MetaTool, channel-keyed interventions induce direction-specific net gains in five models; across three sealed evaluations, none of 59 budget-matched random directions matches calibrated target gain. Crucially, destination auditing shows that behavioral movement does not equal repair: an intervention achieving +55 net gain corrupts over half of the baseline-correct decisions it touches, and promising point estimates on Qwen3-4B and Gemma-2-9B are formally declined due to finite-sample uncertainty. SAKIKO establishes the necessity of outcome-resolved adjudication before claiming internal repair. Code: https://github.com/ruizheliUOA/mechanistic-tool-use-llm.