Cover Image

Where Did It Go Wrong? Attributing Undesirable LLM Behaviors via Representation Gradient Tracing

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Li, Zhe, Zhao, Wei, Li, Yige, Sun, Jun
Format:	Preprint
Published:	2025
Subjects:	Computation and Language Artificial Intelligence Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Zero-Shot Defense Against Toxic Images via Inherent Multimodal Alignment in LVLMs
by: Zhao, Wei, et al.
Published: (2025)

Reducing the Probability of Undesirable Outputs in Language Models Using Probabilistic Inference
by: Zhao, Stephen, et al.
Published: (2025)

CROW: Eliminating Backdoors from Large Language Models via Internal Consistency Regularization
by: Min, Nay Myat, et al.
Published: (2024)

Weak-to-Strong Elicitation via Mismatched Wrong Drafts
by: Deng, Wei
Published: (2026)

RigorLLM: Resilient Guardrails for Large Language Models against Undesired Content
by: Yuan, Zhuowen, et al.
Published: (2024)

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

Do Influence Functions Work on Large Language Models?
by: Li, Zhe, et al.
Published: (2024)

MemTrace: Tracing and Attributing Errors in Large Language Model Memory Systems
by: Deng, Xinle, et al.
Published: (2026)

Probe-Based Data Attribution: Discovering and Mitigating Undesirable Behaviors in LLM Post-Training
by: Xiao, Frank, et al.
Published: (2026)

AttributionBench: How Hard is Automatic Attribution Evaluation?
by: Li, Yifei, et al.
Published: (2024)

R$^2$PO: Decoupling Training Trajectories from Inference Responses for LLM Reasoning
by: Wang, Jingchu, et al.
Published: (2026)

GoRA: Gradient-driven Adaptive Low Rank Adaptation
by: He, Haonan, et al.
Published: (2025)

Unleashing the Unseen: Harnessing Benign Datasets for Jailbreaking Large Language Models
by: Zhao, Wei, et al.
Published: (2024)

BPO: Staying Close to the Behavior LLM Creates Better Online LLM Alignment
by: Xu, Wenda, et al.
Published: (2024)

Easy Problems That LLMs Get Wrong
by: Williams, Sean, et al.
Published: (2024)

ExeChecker: Where Did I Go Wrong?
by: Gu, Yiwen, et al.
Published: (2024)

Precise Attribute Intensity Control in Large Language Models via Targeted Representation Editing
by: Zhang, Rongzhi, et al.
Published: (2025)

Attention Flows: Tracing LLM Conceptual Engagement via Story Summaries
by: Hicke, Rebecca M. M., et al.
Published: (2026)

Tracing the Representation Geometry of Language Models from Pretraining to Post-training
by: Li, Melody Zixuan, et al.
Published: (2025)

GEAR: Granularity-Adaptive Advantage Reweighting for LLM Agents via Self-Distillation
by: Li, Sijia, et al.
Published: (2026)

Rethinking Reward Model Evaluation: Are We Barking up the Wrong Tree?
by: Wen, Xueru, et al.
Published: (2024)

Influential Language Data Selection via Gradient Trajectory Pursuit
by: Deng, Zhiwei, et al.
Published: (2024)

Fantastic Reasoning Behaviors and Where to Find Them: Unsupervised Discovery of the Reasoning Process
by: Zhang, Zhenyu, et al.
Published: (2025)

Towards Real-world Human Behavior Simulation: Benchmarking Large Language Models on Long-horizon, Cross-scenario, Heterogeneous Behavior Traces
by: Chen, Jiawei, et al.
Published: (2026)

Your Simulation Runs but Solves the Wrong Physics: PDE-Grounded Intent Verification for LLM-Generated Multiphysics Simulation Code
by: Song, Zhenghan, et al.
Published: (2026)

Interpreting and Controlling LLM Reasoning through Integrated Policy Gradient
by: Li, Changming, et al.
Published: (2026)

From Noisy Traces to Stable Gradients: Bias-Variance Optimized Preference Optimization for Aligning Large Reasoning Models
by: Zhu, Mingkang, et al.
Published: (2025)

Hidden Heroes and Gradient Bloats: Layer-Wise Redundancy Inverts Attribution in Transformers
by: Ye, Donald
Published: (2026)

Perplexity Cannot Always Tell Right from Wrong
by: Veličković, Petar, et al.
Published: (2026)

GradAlign: Gradient-Aligned Data Selection for LLM Reinforcement Learning
by: Yang, Ningyuan, et al.
Published: (2026)

Rethinking LLM-as-a-Judge: Representation-as-a-Judge with Small Language Models via Semantic Capacity Asymmetry
by: Li, Zhuochun, et al.
Published: (2026)

Watch Every Step! LLM Agent Learning via Iterative Step-Level Process Refinement
by: Xiong, Weimin, et al.
Published: (2024)

Shared Lexical Task Representations Explain Behavioral Variability In LLMs
by: Yang, Zhuonan, et al.
Published: (2026)

Efficient Text-Attributed Graph Learning through Selective Annotation and Graph Alignment
by: Xie, Huanyi, et al.
Published: (2025)

Where Did It Go Wrong? Capability-Oriented Failure Attribution for Vision-and-Language Navigation Agents
by: Chen, Jianming, et al.
Published: (2026)

EmbedLLM: Learning Compact Representations of Large Language Models
by: Zhuang, Richard, et al.
Published: (2024)

Less Is More, but Where? Dynamic Token Compression via LLM-Guided Keyframe Prior
by: Li, Yulin, et al.
Published: (2025)

Hyperdimensional Probe: Decoding LLM Representations via Vector Symbolic Architectures
by: Bronzini, Marco, et al.
Published: (2025)

TruthFlow: Truthful LLM Generation via Representation Flow Correction
by: Wang, Hanyu, et al.
Published: (2025)

Faithfulness Evaluation for Decoder-only LLM Attributions with Controlled Retained Information
by: Huang, Xin, et al.
Published: (2026)