paper-with-me

Papers

The Reasoning Trap: How Enhancing LLM Reasoning Amplifies Tool Hallucination

2025-10-27 · Chenlong Yin, Zeyang Sha, Shiwen Cui, Changhua Meng, Zechao Li arxiv

Enhancing the reasoning capabilities of Large Language Models (LLMs) is a key strategy for building Agents that "think then act." However, recent observations, like OpenAI's o3, suggest a paradox: stronger reasoning often coincides with increased hallucination, yet no prior work has systematically examined whether reasoning enhancement itself causes tool hallucination. To address this gap, we pose the central question: Does strengthening reasoning increase tool hallucination? To answer this, we introduce SimpleToolHalluBench, a diagnostic benchmark measuring tool hallucination in two failure modes: (i) no tool available, and (ii) only distractor tools available. Through controlled experiments, we establish three key findings. First, we demonstrate a causal relationship: progressively enhancing reasoning through RL increases tool hallucination proportionally with task performance gains. Second, this effect transcends overfitting - training on non-tool tasks (e.g., mathematics) still amplifies subsequent tool hallucination. Third, the effect is method-agnostic, appearing when reasoning is instilled via supervised fine-tuning and when it is merely elicited at inference by switching from direct answers to step-by-step thinking. We also evaluate mitigation strategies including Prompt Engineering and Direct Preference Optimization (DPO), revealing a fundamental reliability-capability trade-off: reducing hallucination consistently degrades utility. Mechanistically, Reasoning RL disproportionately collapses tool-reliability-related representations, and hallucinations surface as amplified divergences concentrated in late-layer residual streams. These findings reveal that current reasoning enhancement methods inherently amplify tool hallucination, highlighting the need for new training objectives that jointly optimize for capability and reliability.

📄 PDF Abstract BibTeX arXiv:2510.22977

Code (0)

등록된 구현이 없습니다.

Tasks

Prompt Engineering

Similar Papers 제목 키워드 기반

KnowTrace: Bootstrapping Iterative Retrieval-Augmented Generation with Structured Knowledge Tracing

2025-05-26 · Rui Li, Quanyu Dai, Zeyu Zhang, Xu Chen 외

Recent advances in retrieval-augmented generation (RAG) furnish large language models (LLMs) with iterative retrievals of relevant information to handle complex multi-hop questions. These methods typically alternate betw…

Knowledge TracingMulti-hop Question AnsweringQuestion AnsweringRAG+2

Interleaved Tool-Call Reasoning for Protein Function Understanding

2026-01-07 · Chuanliu Fan, Zicheng Ma, Huanran Meng, Aijia Zhang 외 arxiv

Recent advances in large language models (LLMs) have highlighted the effectiveness of chain-of-thought reasoning in symbolic domains such as mathematics and programming. However, our study shows that directly transferrin…

Protein Function PredictionReinforcement LearningAnswer Generation

Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models

2023-04-23 · Jiashuo Sun, Yi Luo, Yeyun Gong, Chen Lin 외

Large language models (LLMs) can achieve highly effective performance on various reasoning tasks by incorporating step-by-step chain-of-thought (CoT) prompting as demonstrations. However, the reasoning chains of demonstr…

Prototype-Based Dynamic Steering for Large Language Models

2025-10-07 · Ceyhun Efe Kayan, Li Zhang arxiv

Despite impressive breadth, LLMs still rely on explicit reasoning instructions or static, one-fits-all steering methods, leaving a gap for adaptive, instruction-free reasoning amplification. We present Prototype-Based Dy…

Prompt Engineering

OTora: A Unified Red Teaming Framework for Reasoning-Level Denial-of-Service in LLM Agents

2026-05-09 · Xinyu Li, Ronghui Mu, Lin Li, Tianjin Huang 외 arxiv

Large Language Models (LLMs) are increasingly deployed as autonomous agents that execute tool-augmented, multi-step tasks, where latency is a critical factor for real-world applications. Yet an overlooked threat is Reaso…

Red Teaming