Image de couverture de livre

Don't Think Twice! Over-Reasoning Impairs Confidence Calibration

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Lacombe, Romain, Wu, Kerrie, Dilworth, Eddie
Format:	Preprint
Publié:	2025
Sujets:	Artificial Intelligence Computation and Language
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Don't Overthink it. Preferring Shorter Thinking Chains for Improved LLM Reasoning
par: Hassid, Michael, et autres
Publié: (2025)

Reasoning Models Don't Always Say What They Think
par: Chen, Yanda, et autres
Publié: (2025)

Reasoning Models Reason Well, Until They Don't
par: Rameshkumar, Revanth, et autres
Publié: (2025)

Think Twice Before You Write -- an Entropy-based Decoding Strategy to Enhance LLM Reasoning
par: He, Jiashu, et autres
Publié: (2026)

Don't Pay Attention
par: Hammoud, Mohammad, et autres
Publié: (2025)

Don't Think of the White Bear: Ironic Negation in Transformer Models Under Cognitive Load
par: Mann, Logan, et autres
Publié: (2025)

Reasoning about Uncertainty: Do Reasoning Models Know When They Don't Know?
par: Mei, Zhiting, et autres
Publié: (2025)

Don't Trust: Verify -- Grounding LLM Quantitative Reasoning with Autoformalization
par: Zhou, Jin Peng, et autres
Publié: (2024)

Think Twice, Generate Once: Safeguarding by Progressive Self-Reflection
par: Phan, Hoang, et autres
Publié: (2025)

Agentic Confidence Calibration
par: Zhang, Jiaxin, et autres
Publié: (2026)

Don't Forget Your Reward Values: Language Model Alignment via Value-based Calibration
par: Mao, Xin, et autres
Publié: (2024)

Don't Let It Hallucinate: Premise Verification via Retrieval-Augmented Logical Reasoning
par: Qin, Yuehan, et autres
Publié: (2025)

Double-Calibration: Towards Reliable LLMs via Calibrating Knowledge and Reasoning Confidence
par: Lu, Yuyin, et autres
Publié: (2026)

Thinking by Subtraction: Confidence-Driven Contrastive Decoding for LLM Reasoning
par: Tang, Lexiang, et autres
Publié: (2026)

Think Twice: Measuring the Efficiency of Eliminating Prediction Shortcuts of Question Answering Models
par: Mikula, Lukáš, et autres
Publié: (2023)

Larger Language Models Don't Care How You Think: Why Chain-of-Thought Prompting Fails in Subjective Tasks
par: Chochlakis, Georgios, et autres
Publié: (2024)

Reasoning Core: A Scalable RL Environment for LLM Symbolic Reasoning
par: Lacombe, Valentin, et autres
Publié: (2025)

Don't "Overthink" Passage Reranking: Is Reasoning Truly Necessary?
par: Jedidi, Nour, et autres
Publié: (2025)

Align, Don't Divide: Revisiting the LoRA Architecture in Multi-Task Learning
par: Liu, Jinda, et autres
Publié: (2025)

Language Models Don't Learn the Physical Manifestation of Language
par: Lee, Bruce W., et autres
Publié: (2024)

Can AI Assistants Know What They Don't Know?
par: Cheng, Qinyuan, et autres
Publié: (2024)

Confidence-Calibrated Small-Large Language Model Collaboration for Cost-Efficient Reasoning
par: Zhang, Chuang, et autres
Publié: (2026)

Don't Take Things Out of Context: Attention Intervention for Enhancing Chain-of-Thought Reasoning in Large Language Models
par: Yan, Shaotian, et autres
Publié: (2025)

sDPO: Don't Use Your Data All at Once
par: Kim, Dahyun, et autres
Publié: (2024)

VL-Calibration: Decoupled Confidence Calibration for Large Vision-Language Models Reasoning
par: Xiao, Wenyi, et autres
Publié: (2026)

Do Retrieval Augmented Language Models Know When They Don't Know?
par: Zhou, Youchao, et autres
Publié: (2025)

Don't Buy it! Reassessing the Ad Understanding Abilities of Contrastive Multimodal Models
par: Bavaresco, A., et autres
Publié: (2024)

Calibrating Verbalized Confidence with Self-Generated Distractors
par: Wang, Victor, et autres
Publié: (2025)

Fact-Level Confidence Calibration and Self-Correction
par: Yuan, Yige, et autres
Publié: (2024)

Don't Change My View: Ideological Bias Auditing in Large Language Models
par: Kröger, Paul, et autres
Publié: (2025)

Don't Half-listen: Capturing Key-part Information in Continual Instruction Tuning
par: He, Yongquan, et autres
Publié: (2024)

Honest AI: Fine-Tuning "Small" Language Models to Say "I Don't Know", and Reducing Hallucination in RAG
par: Chen, Xinxi, et autres
Publié: (2024)

Show, Don't Tell: Uncovering Implicit Character Portrayal using LLMs
par: Jaipersaud, Brandon, et autres
Publié: (2024)

Think Twice, Click Once: Enhancing GUI Grounding via Fast and Slow Systems
par: Tang, Fei, et autres
Publié: (2025)

Decoupling Reasoning and Confidence: Resurrecting Calibration in Reinforcement Learning from Verifiable Rewards
par: Ma, Zhengzhao, et autres
Publié: (2026)

Why Don't Prompt-Based Fairness Metrics Correlate?
par: Zayed, Abdelrahman, et autres
Publié: (2024)

You Don't Need Prompt Engineering Anymore: The Prompting Inversion
par: Khan, Imran
Publié: (2025)

s3: You Don't Need That Much Data to Train a Search Agent via RL
par: Jiang, Pengcheng, et autres
Publié: (2025)

Don't Let It Fade: Preserving Edits in Diffusion Language Models via Token Timestep Allocation
par: Kim, Woojin, et autres
Publié: (2025)

Can LLMs Ground when they (Don't) Know: A Study on Direct and Loaded Political Questions
par: Lachenmaier, Clara, et autres
Publié: (2025)