Imagen de Portada

Navigating by Old Maps: The Pitfalls of Static Mechanistic Localization in LLM Post-Training

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Chen, Hang, Zhu, Jiaying, Chen, Hongyang, Liu, Hongxu, Yang, Xinyu, Wang, Wenya
Formato:	Preprint
Publicado:	2026
Materias:	Computation and Language
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

CLUE: Conflict-guided Localization for LLM Unlearning Framework
por: Chen, Hang, et al.
Publicado: (2025)

Quantifying Semantic Emergence in Language Models
por: Chen, Hang, et al.
Publicado: (2024)

Skill Path: Unveiling Language Skills from Circuit Graphs
por: Chen, Hang, et al.
Publicado: (2024)

From Hallucination to Structure Snowballing: The Alignment Tax of Constrained Decoding in LLM Reflection
por: Zhou, Hongxu
Publicado: (2026)

Rethinking Circuit Completeness in Language Models: AND, OR, and ADDER Gates
por: Chen, Hang, et al.
Publicado: (2025)

Tracking the Feature Dynamics in LLM Training: A Mechanistic Study
por: Xu, Yang, et al.
Publicado: (2024)

Rethinking Local Learning: A Cheaper and Faster Recipe for LLM Post-Training
por: Shi, Hengyu, et al.
Publicado: (2026)

Task-Aware LLM Routing with Multi-Level Task-Profile-Guided Data Synthesis for Cold-Start Scenarios
por: Liu, Hui, et al.
Publicado: (2026)

Mechanistic Data Attribution: Tracing the Training Origins of Interpretable LLM Units
por: Chen, Jianhui, et al.
Publicado: (2026)

Reinforcement Learning for LLM Post-Training: A Survey
por: Wang, Zhichao, et al.
Publicado: (2024)

Right Is Not Enough: The Pitfalls of Outcome Supervision in Training LLMs for Math Reasoning
por: Guo, Jiaxing, et al.
Publicado: (2025)

Consolidating Rewarded Perturbations for LLM Post-Training
por: Zhang, Zheyu, et al.
Publicado: (2026)

Learning a Structural Causal Model for Intuition Reasoning in Conversation
por: Chen, Hang, et al.
Publicado: (2023)

Training Language Models to Generate Text with Citations via Fine-grained Rewards
por: Huang, Chengyu, et al.
Publicado: (2024)

Pitfalls and Outlooks in Using COMET
por: Zouhar, Vilém, et al.
Publicado: (2024)

Can LLMs Learn to Map the World from Local Descriptions?
por: Xia, Sirui, et al.
Publicado: (2025)

Can Post-Training Quantization Benefit from an Additional QLoRA Integration?
por: Zhu, Xiliang, et al.
Publicado: (2025)

Examining Reasoning LLMs-as-Judges in Non-Verifiable LLM Post-Training
por: Liu, Yixin, et al.
Publicado: (2026)

Mind the Gap: Pitfalls of LLM Alignment with Asian Public Opinion
por: Shankar, Hari, et al.
Publicado: (2026)

Joint Effects of Argumentation Theory, Audio Modality and Data Enrichment on LLM-Based Fallacy Classification
por: Zhou, Hongxu, et al.
Publicado: (2025)

AdaDecode: Accelerating LLM Decoding with Adaptive Layer Parallelism
por: Wei, Zhepei, et al.
Publicado: (2025)

Prompt Curriculum Learning for Efficient LLM Post-Training
por: Gao, Zhaolin, et al.
Publicado: (2025)

Narrative Flattening: How Post-Training Compresses Thematic, Affective, and Stylistic Variation in LLM Fiction
por: Li, Zehan, et al.
Publicado: (2026)

Why Does Reinforcement Learning Generalize? A Feature-Level Mechanistic Study of Post-Training in Large Language Models
por: Shi, Dan, et al.
Publicado: (2026)

Large Language Models Know What Makes Exemplary Contexts
por: Long, Quanyu, et al.
Publicado: (2024)

Post-training an LLM for RAG? Train on Self-Generated Demonstrations
por: Finlayson, Matthew, et al.
Publicado: (2025)

LLM-based Human Simulations Have Not Yet Been Reliable
por: Wang, Qian, et al.
Publicado: (2025)

The UNDO Flip-Flop: A Controlled Probe for Reversible Semantic State Management in State Space Model
por: Zhou, Hongxu
Publicado: (2026)

Alternating Reinforcement Learning for Rubric-Based Reward Modeling in Non-Verifiable LLM Post-Training
por: Xu, Ran, et al.
Publicado: (2026)

Don't Get Lost in the Trees: Streamlining LLM Reasoning by Overcoming Tree Search Exploration Pitfalls
por: Wang, Ante, et al.
Publicado: (2025)

The Pitfalls of Defining Hallucination
por: van Deemter, Kees
Publicado: (2024)

Beware of Reasoning Overconfidence: Pitfalls in the Reasoning Process for Multi-solution Tasks
por: Guan, Jiannan, et al.
Publicado: (2025)

A Survey on LLM Mid-Training
por: Tu, Chengying, et al.
Publicado: (2025)

Towards a Mechanistic Understanding of Large Reasoning Models: A Survey of Training, Inference, and Failures
por: Hu, Yi, et al.
Publicado: (2026)

Can Language Models Act as Knowledge Bases at Scale?
por: He, Qiyuan, et al.
Publicado: (2024)

SASQ: Static Activation Scaling for Quantization-Aware Training in Large Language Models
por: Mao, Shizhuo, et al.
Publicado: (2025)

Do LLM Evaluators Prefer Themselves for a Reason?
por: Chen, Wei-Lin, et al.
Publicado: (2025)

Reinforcing Compositional Retrieval: Retrieving Step-by-Step for Composing Informative Contexts
por: Long, Quanyu, et al.
Publicado: (2025)

Beyond Static Cropping: Layer-Adaptive Visual Localization and Decoding Enhancement
por: Zhu, Zipeng, et al.
Publicado: (2026)

PPA-Plan: Proactive Pitfall Avoidance for Reliable Planning in Long-Context LLM Reasoning
por: Kim, Byeongjin, et al.
Publicado: (2026)