mlsanityData to 14 Oct 2026

When Tools Hurt LLM Reasoning: State-Dependent Belief Revision under External Evidence

arXiv 2508.15754 ↗First posted 21 Aug 2025cs.CL cs.AIcited by 3

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

Tool use is often assumed to monotonically improve reasoning, where external evidence is expected to help when relevant and be ignored when irrelevant. We show that this assumption fails in a state-dependent way. Across benchmarks with Python and Wikipedia tools, external evidence reliably helps when initial beliefs are weak, but can flip already-correct answers when those beliefs are strong. We frame this as a misallocation of revision authority, arguing that deferring to external evidence is suboptimal when internal support for the correct answer surpasses the tool's expected output quality. This predicts that harm should concentrate on high-confidence no-tool cases. We test this prediction with threshold localization, wrong-trace audits, and a same-clue intervention showing that revision framing changes the damage caused by misleading evidence. These findings suggest that mixed no-tool/tool-assisted inference should arbitrate authority rather than privilege tool evidence by default. As a minimal demonstration, we introduce CASE, a label-free controller that selects between no-tool and tool-assisted trajectories using answer-state certainty and improves over existing confidence-based arbitration baselines.The code for our experiments is available at https://github.com/epsilondylan/State-Dependent-Belief-Revision.

Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 7 of its 49 references that are in the map. First posted on arXiv 21 Aug 2025.

Builds on

7 of 49 references.

References · 7

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2022Language Models (Mostly) Know What They Knowcited by 1,856
  3. 2025Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learningcited by 1,444
  4. 2022Solving math word problems with process- and outcome-based feedbackcited by 901
  5. 2023FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performancecited by 853
  6. 2025ReTool: Reinforcement Learning for Strategic Tool Use in LLMscited by 392
  7. 2023ART: Automatic multi-step reasoning and tool-use for large language modelscited by 197

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026Beyond Accuracy: Unveiling Inefficiency Patterns in Tool-Integrated Reasoningcited by 3
  2. 2026ReaComp: Compiling LLM Reasoning into Symbolic Solvers for Efficient Program Synthesiscited by 2
  3. 2026TIDE-Bench: Task-Aware and Diagnostic Evaluation of Tool-Integrated Reasoningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Don't Blindly Trust It: How Unreliable Feedback Breaks Tool-Using LLM Agents
  2. 2026When Tools Get in the Way: The Effect of Unnecessary Tool Availability on LLM Answering
  3. 2026The Tool-Overuse Illusion: Why Does LLM Prefer External Tools over Internal Knowledge?
  4. 2026Judges Follow the Reference: The Derivation Gap in Verifying Tool-Call Arguments
  5. 2026Easier to Mislead Than to Correct: Harmful and Beneficial Revision in LLM Conformity
  6. 2026DeltaLogic: Minimal Premise Edits Reveal Belief-Revision Failures in Logical Reasoning Models
  7. 2025From Proof to Program: Characterizing Tool-Induced Reasoning Hallucinations in Large Language Models
  8. 2026When Evidence Changes: Evaluating Memory Repair and Re-reading in Language-Model Agents