Looking Is Not Picking: An Attention-Segment Account of Tool-Selection Failures in LLM Agents

• URL: http://arxiv.org/abs/2606.16364v1
• Date: Mon, 15 Jun 2026 07:58:56 GMT
• Title: Looking Is Not Picking: An Attention-Segment Account of Tool-Selection Failures in LLM Agents
• Abstract summary: LLM agents mis-call tools, and the natural guess is that the model failed to see the right tool in a crowded harness.<n>We show the opposite through a lens concurrent work sets aside -- the model's attention to labeled tool-definition segments.
• Score: 0.4855342112648282
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: LLM agents mis-call tools, and the natural guess is that the model failed to see the right tool in a crowded harness. We show the opposite through a lens concurrent work sets aside -- the model's attention to labeled tool-definition segments. On real BFCL failures, by per-candidate attention argmax the model attends most to the correct tool 80% of the time (vs. 21% chance), and the gold is the under-attended segment on only 10%: it looks at the right tool and still picks wrong. This directly refutes the intuitive "crowded-harness / lost-in-the-middle" explanation: the failure is at the decision readout, not the harness, and we pin it there three ways. (1) Input vs. readout: repairing the prompt (reordering or duplicating the gold tool) recovers <=23% of failures, while readout-side interventions recover 59-91%. (2) Representation-invariance: two gold-pointed interventions in different representations -- an additive attention-logit bias and a residual-stream steering vector -- recover largely the same failures (per-task Jaccard 0.865 pooled, 0.79-0.91 per model), so the bottleneck is localized to the readout independent of which representation is poked. (3) A training-free, gold-free selector: per-segment attention closes most of the gold-free-vs-oracle gap on BFCL (+11.9 pts pooled function-name selection vs. +17.9-pt oracle headroom) and adds +14.9 pts on Seal-Tools; every model positive (exact McNemar p<=8e-4 each). Scopes differ: the causal attention-bias dose-response is bidirectional and monotonic on 10 mask-honoring models (3-32B), the full 0.5-32B span carrying only the correlational diagnostic; the deployable selector is evaluated on 5 single-turn models and does not yet transfer to a multi-turn loop.

Related papers

• ReDraft, Don't Just Distill: Reference-Driven Revision for Continual VLLM Post-Training [66.72708893922605]
  Post-training of large multimodal models should add new capabilities while preserving those from pre-training.<n>SFT gives explicit target supervision that learns a task from near-zero accuracy, but its off-policy targets move the model far enough to cause forgetting.<n>We introduce ReDraft, which obtains both from the model's own failures.
  arXiv  Detail & Related papers  (2026-09-15T04:59:03Z)
• When Do Internal Probes Beat Reading the Answer? Miscalibrated Readouts and Behavior-Concealed Knowledge in Language Models [0.0]
  A language model, asked to verify 1,200 logical conclusions, answers YES every time.<n>linear probes on its hidden states read the correct verdict at 0.96 AUC.<n>We ask where the verdict is lost, and find the dominant failure is a single scalar.<n>The diagnosis generalizes: across 90 semantic-label configurations of a five-model, three-family factorial, behavioral accuracy collapses onto a single function of threshold offset.
  arXiv  Detail & Related papers  (2026-09-04T00:26:45Z)
• Calibration is the Bottleneck: An Action-Class Diagnostic of Multi-Turn Tool-Calling [65.03665015319457]
  We propose an action-class-oriented diagnostic framework that decomposes multi-turn failures into two modes: action-class miscalibration and action-execution failure.<n>We validate it on a panel of tool-calling models across multiple multi-turn benchmarks.
  arXiv  Detail & Related papers  (2026-09-01T09:08:15Z)
• Feedback That Backfires: Why Small Language Model Agents Repeat the Call They Just Watched Fail [0.0]
  Defining the corrective gain of a failure record as the change in log-probability of re-emitting the action that just failed, we find the gain is negative for every instruction-tuned model.<n>The problem is in the harness, not the model's grasp of error messages, and that predicts which remedies work.
  arXiv  Detail & Related papers  (2026-08-24T10:19:17Z)
• DarwinX: Evolving Agent Harnesses Through Natural Selection [48.64258219266629]
  DarwinX treats self-evolution as selection over a population of harnesses with the model frozen.<n>General agent competence, not benchmark-specific patches, so it survives changes of task, verifier, and base model.
  arXiv  Detail & Related papers  (2026-07-31T05:34:02Z)
• Interaction Scaling: Grounding the Third Axis of Test-Time Compute [0.33451037881913753]
  We argue that a single variable governs this third axis, grounding, and that it must hold on both sides of the loop.<n>On hard coding tasks at a fixed token budget, reasoning-only and best-of-N sampling both plateau.<n>Our proposer-reviewer harness reaches a perfect 100% pass rate with no run-to-run variance.
  arXiv  Detail & Related papers  (2026-07-13T14:22:06Z)
• When the Tool Decides: LLM Agents Defer Blindly to Graph Neural Network Tools, and Stronger Backbones Defer More [10.283803346813047]
  A growing line of work equips large language model (LLM) agents with graph neural networks (GNNs) as callable tools.<n>We expose a frozen GNN to a ReAct-style LLM agent as an explicit tool and measure, on node classification over a text-attributed graph.<n>We find the agent does not exercise judgment: its predictions agree with the raw GNN's 97.6-99.2% of the time (5 seeds)<n>We find the cost of deference does not shrink as capability grows and grows where alternatives emerge.
  arXiv  Detail & Related papers  (2026-06-12T14:13:19Z)
• PURGE: Projected Unlearning via Retain-Guided Erasure [5.069332646113575]
  PURGE is a machine unlearning algorithm built on a simple but an under-exploited observation: continual learning (CL) and machine unlearning (MU)<n> CL tries to learn new tasks without forgetting old ones; MU tries to erase specific data without hurting retained performance representing the same underlying tension in opposite directions.<n> PURGE consistently keeps accuracy above 96% while achieving MIA AUROC close to 0.5 (the ideal), outperforming gradient ascent, KL-uniform, and several published baselines on the privacy-utility frontier.
  arXiv  Detail & Related papers  (2026-06-02T15:53:01Z)
• Right Makes Might: Aligning Verified Hidden States Empowers RL Reasoning [55.264863369127774]
  Current methods reduce each correct rollout to a single reward bit, ignoring the geometric structure shared among their hidden states.<n>We propose Hidden-Align, an auxiliary loss function that aligns the last-layer hidden states of correct rollouts at the anchor token during RL training, with zero overhead in both training and inference.<n>On eight mathematical reasoning benchmarks, Hidden-Align improves average pass@1 over the DAPO baseline by 3.8, 6.2, and 5.4 percentage points on Qwen3-1.7B, 4B, and 14B respectively, with consistent pass@k gains across all three scales.
  arXiv  Detail & Related papers  (2026-06-02T06:51:15Z)
• Self-Healing Agentic Orchestrators for Reliable Tool-Augmented Large Language Model Systems [0.10210859604701106]
  Self-healing agentic orchestrator treats reliability as a runtime bounded control problem.<n>Self-healing achieves 98.8% task success, compared with 94.5% for retry-only and 93.8% for full replanning.
  arXiv  Detail & Related papers  (2026-05-31T19:27:22Z)
• GCE-MIL: Faithful and Recoverable Evidence for Multiple Instance Learning in Whole-Slide Imaging [4.001315295826918]
  Multiple instance learning (MIL) is the standard approach for whole-slide image (WSI) classification and survival prediction.<n>GCE-MIL is a backbone-agnostic wrapper implemented through three injection modes and three evidence components.
  arXiv  Detail & Related papers  (2026-05-17T13:56:30Z)
• Tool Calling is Linearly Readable and Steerable in Language Models [7.851713181228514]
  When a tool-calling agent picks the wrong tool, the failure is invisible until execution.<n>We use 12 instruction-tuned models across Gemma 3, Qwen 3, Qwen 2.5, and Llama 3.1.<n>We find the identity of the chosen tool is linearly and steerable inside the model.
  arXiv  Detail & Related papers  (2026-05-08T16:47:08Z)
• Pando: Do Interpretability Methods Work When Models Won't Explain Themselves? [53.07826484214082]
  We introduce Pando, a model-organism benchmark that breaks this confound via an explanation axis.<n>Pando predicts held-out model decisions from $10$ labeled query-response pairs.<n>When explanations are faithful, black-box elicitation matches or exceeds all white-box methods.
  arXiv  Detail & Related papers  (2026-04-13T06:42:24Z)
• CARE What Fails: Contrastive Anchored-REflection for Verifiable Multimodal [84.71254539482369]
  Group-relative reinforcement learning with verifiable rewards (RLVR) often wastes the most informative data it already has the failures.<n>We present CARE, a failure-centric post-training framework for multimodal reasoning that turns errors into supervision.<n> CARE improves accuracy and training smoothness while explicitly increasing the share of learning signal that comes from failures.
  arXiv  Detail & Related papers  (2025-12-22T16:34:21Z)
• SEAL: Steerable Reasoning Calibration of Large Language Models for Free [58.931194824519935]
  Large Language Models (LLMs) have demonstrated compelling capabilities for complex reasoning tasks via the extended chain-of-thought (CoT) reasoning mechanism.<n>Recent studies reveal substantial redundancy in the CoT reasoning traces, which negatively impacts model performance.<n>We introduce SEAL, a training-free approach that seamlessly calibrates the CoT process, improving accuracy while demonstrating significant efficiency gains.
  arXiv  Detail & Related papers  (2025-04-07T02:42:07Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.