REVIEW 30 cited by
The Danger of Overthinking: Examining the Reasoning-Action Dilemma in Agentic Tasks
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
The Danger of Overthinking: Examining the Reasoning-Action Dilemma in Agentic Tasks
read the original abstract
Large Reasoning Models (LRMs) represent a breakthrough in AI problem-solving capabilities, but their effectiveness in interactive environments can be limited. This paper introduces and analyzes overthinking in LRMs. A phenomenon where models favor extended internal reasoning chains over environmental interaction. Through experiments on software engineering tasks using SWE Bench Verified, we observe three recurring patterns: Analysis Paralysis, Rogue Actions, and Premature Disengagement. We propose a framework to study these behaviors, which correlates with human expert assessments, and analyze 4018 trajectories. We observe that higher overthinking scores correlate with decreased performance, with reasoning models exhibiting stronger tendencies toward overthinking compared to non-reasoning models. Our analysis reveals that simple efforts to mitigate overthinking in agentic environments, such as selecting the solution with the lower overthinking score, can improve model performance by almost 30% while reducing computational costs by 43%. These results suggest that mitigating overthinking has strong practical implications. We suggest that by leveraging native function-calling capabilities and selective reinforcement learning overthinking tendencies could be mitigated. We also open-source our evaluation framework and dataset to facilitate research in this direction at https://github.com/AlexCuadron/Overthinking.
Forward citations
Cited by 30 Pith papers
-
TRIAGE: Evaluating Prospective Metacognitive Control in LLMs under Resource Constraints
TRIAGE evaluates LLMs on prospective metacognitive control by requiring a single plan for task selection, sequencing, and token allocation under a calibrated budget, revealing substantial gaps in current models across...
-
AcademiClaw: When Students Set Challenges for AI Agents
AcademiClaw is a new benchmark of 80 student-sourced academic tasks where the best frontier AI agents achieve only a 55% pass rate.
-
Benchmarking and Evaluating VLMs for Software Architecture Diagram Understanding
SADU benchmark shows top VLMs reach only 70% accuracy on software architecture diagram tasks, revealing gaps in visual reasoning for engineering artifacts.
-
Diagnosing Search Behavior and Failure Modes in Long-Horizon Search Agents
For six deep search agents on BrowseComp-Plus, answer accuracy tracks cumulative retrieval recall, not search effort, and failures split into missing-evidence and evidence-misuse gaps.
-
Scores Are Not Decisions: Cost-Aware Stopping for Tool Acquisition in LLM Agents
Cost-aware stopping over ranked tool prefixes, trained on stop-versus-continue payoff gaps, beats score and sufficiency thresholds under heterogeneous tool costs.
-
EvoThink: Evolving Thinking in Large Reasoning Models via Self-Pruning and Aha-Moment Preference Optimization
EvoThink reduces LRM overthinking by pruning redundant atomic reasoning steps (SPT) and training on diversity-selected wrong-to-right mutation data (AMPO), cutting tokens and improving math/code accuracy.
-
Implicit Fine-tuning via Context Engineering: A Curriculum Learning Framework for Multimodal Entity Alignment
Context engineering for multimodal entity alignment is mathematically equivalent to sequential contrastive fine-tuning, enabling a curriculum prompt framework that matches large-model accuracy at far lower cost.
-
From Trajectories to Prefixes: Reusing Teacher Trajectories via Replayed Prefixes and Online Continuation
Replaying teacher trajectory prefixes and explicitly optimizing the historical prefix tokens improves small-model agent success rates over distillation and response-only GRPO baselines in TextCraft, BabyAI, and ALFWorld.
-
SQLConductor: Search-to-Policy Learning for Step-wise Text-to-SQL Orchestration
SQLConductor uses Search-to-Policy Learning with MCTS, stability-weighted SFT, and curriculum RL to train a compact policy for adaptive step-wise Text-to-SQL orchestration, reporting 73.2% EX on BIRD-Dev.
-
Dynamic Rollout Editing for Reducing Overthinking in RL-Trained Reasoning Models
Dynamic Rollout Editing reduces overthinking in RL-trained LLMs by editing post-answer continuations in successful rollouts and preferring the edited versions within GRPO groups.
-
Human oversight of agentic systems in practice: Examining the oversight work, challenges, and heuristics of developers using software agents
Exploratory interview study with 17 developers identifies four forms of emergent oversight work for software agents and documents situated challenges and heuristics.
-
Thinking Past the Answer: Evaluating Harmful Overthinking in Large Reasoning Models
Stopping large reasoning models at the first correct reasoning prefix improves accuracy up to 21% by avoiding harmful overthinking that destabilizes correct trajectories.
-
Inducing Overthink: Hierarchical Genetic Algorithm-based DoS Attack on Black-Box Large Language Reasoning Models
A hierarchical genetic algorithm induces overthinking in black-box large reasoning models by perturbing logical structure, achieving up to 26.1x longer outputs on the MATH benchmark.
-
Inducing Overthink: Hierarchical Genetic Algorithm-based DoS Attack on Black-Box Large Language Reasoning Models
A hierarchical genetic algorithm induces overthinking in black-box LRMs, increasing output length by up to 26.1x on the MATH benchmark.
-
Implicit Compression Regularization: Concise Reasoning via Internal Shorter Distributions in RL Post-Training
ICR creates a virtual shorter distribution from shortest correct on-policy responses to regularize RL post-training toward concise yet accurate reasoning, improving the accuracy-length Pareto frontier on math and know...
-
FitText: Evolving Agent Tool Ecologies via Memetic Retrieval
FitText embeds evolutionary retrieval of tool descriptions into the agent loop, yielding 2.7-10.6 point NDCG@5 gains on ToolRet and 26.7-point pass-rate gains on StableToolBench.
-
FitText: Evolving Agent Tool Ecologies via Memetic Retrieval
Agents that evolve pseudo-tool search descriptions during execution reach an 84.3% StableToolBench pass rate, +26.7 points over static query retrieval, with gains growing with base-model capability.
-
FitText: Evolving Agent Tool Ecologies via Memetic Retrieval
FitText embeds memetic evolutionary retrieval inside the agent's reasoning loop to iteratively refine pseudo-tool descriptions, raising retrieval rank from 8.81 to 2.78 on ToolRet and pass rate to 0.73 on StableToolBench.
-
Reasoning While Asking: Transforming Reasoning Large Language Models from Passive Solvers to Proactive Inquirers
Proactive Interactive Reasoning (PIR) teaches LLMs to insert clarification questions into their chain-of-thought, improving simulated task success while cutting reasoning tokens roughly in half.
-
Understanding the Effects of Distractors on Reasoning Vision-Language Models
Adding visual distractors to images lowers reasoning-VLM accuracy while leaving reasoning length essentially flat, with the accuracy loss traced to the share of reasoning-trace attributes devoted to distractors.
-
TeaRAG: A Token-Efficient Agentic Retrieval-Augmented Generation Framework
TeaRAG shows that hybrid chunk+triplet retrieval with Personalized PageRank and an iterative process-aware DPO reward keeps QA accuracy while cutting reasoning tokens by roughly 60%.
-
Efficient Test-Time Scaling via Temporal Reasoning Aggregation
TRACE aggregates answer consistency and confidence trajectory over multiple reasoning steps to decide when to halt inference, reducing token usage by 25-30% while keeping accuracy within 1-2% of full reasoning.
-
Adapt to Thrive! Adaptive Power-Mean Policy Optimization for Improved LLM Reasoning
APMPO boosts average Pass@1 scores on math reasoning benchmarks by 3 points over GRPO by using an adaptive power-mean policy objective and feedback-driven clipping bounds in RLVR training.
-
Free Energy-Driven Reinforcement Learning with Adaptive Advantage Shaping for Unsupervised Reasoning in LLMs
FREIA applies free energy principles and adaptive advantage shaping to unsupervised RL, outperforming baselines by 0.5-3.5 Pass@1 points on math reasoning with a 1.5B model.
-
When Is Thinking Enough? Early Exit via Sufficiency Assessment for Efficient Reasoning
DTSR enables large reasoning models to dynamically assess chain-of-thought sufficiency via reflection signals and a sufficiency check, reducing reasoning length by 28.9-34.9% with minimal performance loss on Qwen3 models.
-
Large Language Models as Virtual Survey Respondents: Evaluating Sociodemographic Response Generation
Introduces PAS and FAS task abstractions plus the LLM-S^3 benchmark to evaluate LLMs on generating sociodemographic survey responses across 11 real datasets and multiple models.
-
Self-Aligned Reward: Towards Effective and Efficient Reasoners
Self-aligned reward uses relative perplexity differences to encourage concise, query-specific reasoning in LLMs, yielding 4% accuracy gains and 30% lower inference cost when added to PPO or GRPO.
-
Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models
The paper unifies perspectives on Long CoT in reasoning LLMs by introducing a taxonomy, detailing characteristics of deep reasoning and reflection, and discussing emergence phenomena and future directions.
-
The Periodic Table of LLM Reasoning: A Structured Survey of Reasoning Paradigms, Methods, and Failure Modes
A literature survey that introduces a taxonomy for LLM reasoning paradigms, analyzes methodological trends, and synthesizes failure modes from over 300 papers.
-
When Researchers Say Mental Model/Theory of Mind of AI, What Are They Really Talking About?
Researchers' claims of AI theory of mind are really about behavioral prediction, so AI evaluation should shift from isolated cognitive tests to human-AI interaction.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.