Imagen de Portada

Can We Predict Alignment Before Models Finish Thinking? Towards Monitoring Misaligned Reasoning Models

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Chan, Yik Siu, Yong, Zheng-Xin, Bach, Stephen H.
Formato:	Preprint
Publicado:	2025
Materias:	Computation and Language Artificial Intelligence Machine Learning
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

LexC-Gen: Generating Data for Extremely Low-Resource Languages with Large Language Models and Bilingual Lexicons
por: Yong, Zheng-Xin, et al.
Publicado: (2024)

Can We Predict Before Executing Machine Learning Agents?
por: Zheng, Jingsheng, et al.
Publicado: (2026)

AdaptThink: Reasoning Models Can Learn When to Think
por: Zhang, Jiajie, et al.
Publicado: (2025)

Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking
por: Zelikman, Eric, et al.
Publicado: (2024)

Think Before You Prune: Self-Reflective Structured Pruning for Reasoning Language Models
por: Wang, Ziyan, et al.
Publicado: (2025)

Think Before You Lie: How Reasoning Leads to Honesty
por: Yuan, Ann, et al.
Publicado: (2026)

Speak Easy: Eliciting Harmful Jailbreaks from LLMs with Simple Interactions
por: Chan, Yik Siu, et al.
Publicado: (2025)

Low-Resource Languages Jailbreak GPT-4
por: Yong, Zheng-Xin, et al.
Publicado: (2023)

Preference Tuning For Toxicity Mitigation Generalizes Across Languages
por: Li, Xiaochen, et al.
Publicado: (2024)

Thought Crime: Backdoors and Emergent Misalignment in Reasoning Models
por: Chua, James, et al.
Publicado: (2025)

Ulterior Motives: Detecting Misaligned Reasoning in Continuous Thought Models
por: Ramjee, Sharan
Publicado: (2026)

Dynamic Thinking-Token Selection for Efficient Reasoning in Large Reasoning Models
por: Guo, Zhenyuan, et al.
Publicado: (2026)

Self-Jailbreaking: Language Models Can Reason Themselves Out of Safety Alignment After Benign Reasoning Training
por: Yong, Zheng-Xin, et al.
Publicado: (2025)

MeTHanol: Modularized Thinking Language Models with Intermediate Layer Thinking, Decoding and Bootstrapping Reasoning
por: Xi, Ningyuan, et al.
Publicado: (2024)

To Think or Not to Think: Exploring the Unthinking Vulnerability in Large Reasoning Models
por: Zhu, Zihao, et al.
Publicado: (2025)

AdapThink: Adaptive Thinking Preferences for Reasoning Language Model
por: Wan, Xu, et al.
Publicado: (2025)

Reasoning Models Can Be Effective Without Thinking
por: Ma, Wenjie, et al.
Publicado: (2025)

Think Before you Write: QA-Guided Reasoning for Character Descriptions in Books
por: Papoudakis, Argyrios, et al.
Publicado: (2026)

Crosslingual Reasoning through Test-Time Scaling
por: Yong, Zheng-Xin, et al.
Publicado: (2025)

Efficient Reasoning with Balanced Thinking
por: Li, Yulin, et al.
Publicado: (2026)

Effectively Controlling Reasoning Models through Thinking Intervention
por: Wu, Tong, et al.
Publicado: (2025)

Reasoning Models Don't Always Say What They Think
por: Chen, Yanda, et al.
Publicado: (2025)

Think Before You Act: Decision Transformers with Working Memory
por: Kang, Jikun, et al.
Publicado: (2023)

Thinking About Thinking: SAGE-nano's Inverse Reasoning for Self-Aware Language Models
por: Jha, Basab, et al.
Publicado: (2025)

BRiTE: Bootstrapping Reinforced Thinking Process to Enhance Language Model Reasoning
por: Zhong, Han, et al.
Publicado: (2025)

The Detection-Extraction Gap: Models Know the Answer Before They Can Say It
por: Wang, Hanyang, et al.
Publicado: (2026)

Epistemic Traps: Rational Misalignment Driven by Model Misspecification
por: Xu, Xingcheng, et al.
Publicado: (2026)

Metaphors are a Source of Cross-Domain Misalignment of Large Reasoning Models
por: Hu, Zhibo, et al.
Publicado: (2026)

Can Large Language Models Reason and Plan?
por: Kambhampati, Subbarao
Publicado: (2024)

ThinkRouter: Efficient Reasoning via Routing Thinking between Latent and Discrete Spaces
por: Xu, Xin, et al.
Publicado: (2026)

Slow-Fast Policy Optimization: Reposition-Before-Update for LLM Reasoning
por: Wang, Ziyan, et al.
Publicado: (2025)

Demystifying Hybrid Thinking: Can LLMs Truly Switch Between Think and No-Think?
por: Wang, Shouren, et al.
Publicado: (2025)

Reasoning with Sampling: Your Base Model is Smarter Than You Think
por: Karan, Aayush, et al.
Publicado: (2025)

The Thinking Spectrum: An Empirical Study of Tunable Reasoning in LLMs through Model Merging
por: Lan, Xiaochong, et al.
Publicado: (2025)

Re-Emergent Misalignment: How Narrow Fine-Tuning Erodes Safety Alignment in LLMs
por: Giordani, Jeremiah
Publicado: (2025)

Masked Thought: Simply Masking Partial Reasoning Steps Can Improve Mathematical Reasoning Learning of Language Models
por: Chen, Changyu, et al.
Publicado: (2024)

Persona-Model Collapse in Emergent Misalignment
por: Costa, Davi Bastos, et al.
Publicado: (2026)

Missing Premise exacerbates Overthinking: Are Reasoning Models losing Critical Thinking Skill?
por: Fan, Chenrui, et al.
Publicado: (2025)

More Thinking, More Bias: Length-Driven Position Bias in Reasoning Models
por: Wang, Xiao
Publicado: (2026)

Dyna-Think: Synergizing Reasoning, Acting, and World Model Simulation in AI Agents
por: Yu, Xiao, et al.
Publicado: (2025)