Cover Image

LLM Knowledge is Brittle: Truthfulness Representations Rely on Superficial Resemblance

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Haller, Patrick, Ibrahim, Mark, Kirichenko, Polina, Sagun, Levent, Bell, Samuel J.
Format:	Preprint
Published:	2025
Subjects:	Computation and Language Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

An Effective Theory of Bias Amplification
by: Subramonian, Arjun, et al.
Published: (2024)

Reassessing the Validity of Spurious Correlations Benchmarks
by: Bell, Samuel J., et al.
Published: (2024)

Modeling Caption Diversity in Contrastive Vision-Language Pretraining
by: Lavoie, Samuel, et al.
Published: (2024)

From Data to Knowledge: Evaluating How Efficiently Language Models Learn Facts
by: Christoph, Daniel, et al.
Published: (2025)

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

Verify with Caution: The Pitfalls of Relying on Imperfect Factuality Metrics
by: Godbole, Ameya, et al.
Published: (2025)

Feature Resemblance: Towards a Theoretical Understanding of Analogical Reasoning in Transformers
by: Xu, Ruichen, et al.
Published: (2026)

What's in Common? Multimodal Models Hallucinate When Reasoning Across Scenes
by: Ross, Candace, et al.
Published: (2025)

On the Role of Speech Data in Reducing Toxicity Detection Bias
by: Bell, Samuel J., et al.
Published: (2024)

Non-Linear Inference Time Intervention: Improving LLM Truthfulness
by: Hoscilowicz, Jakub, et al.
Published: (2024)

Knowledge Distillation in Automated Annotation: Supervised Text Classification with LLM-Generated Training Labels
by: Pangakis, Nicholas, et al.
Published: (2024)

Pruning Weights but Not Truth: Safeguarding Truthfulness While Pruning LLMs
by: Fu, Yao, et al.
Published: (2025)

Chained Tuning Leads to Biased Forgetting
by: Ung, Megan, et al.
Published: (2024)

Networked Inequality: Preferential Attachment Bias in Graph Neural Network Link Prediction
by: Subramonian, Arjun, et al.
Published: (2023)

Truth as a Trajectory: What Internal Representations Reveal About Large Language Model Reasoning
by: Damirchi, Hamed, et al.
Published: (2026)

Prompt Attacks Reveal Superficial Knowledge Removal in Unlearning Methods
by: Jang, Yeonwoo, et al.
Published: (2025)

Position: The Turing-Completeness of Autoregressive Transformers Relies Heavily on Context Management
by: Cui, Guanyu, et al.
Published: (2026)

Revisiting the Superficial Alignment Hypothesis
by: Raghavendra, Mohit, et al.
Published: (2024)

Beyond Simple Averaging: Improving NLP Ensemble Performance with Topological-Data-Analysis-Based Weighting
by: Proskura, Polina, et al.
Published: (2024)

Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications
by: Wei, Boyi, et al.
Published: (2024)

Automated Knowledge Concept Annotation and Question Representation Learning for Knowledge Tracing
by: Ozyurt, Yilmazcan, et al.
Published: (2024)

The Truth Lies Somewhere in the Middle (of the Generated Tokens)
by: Wang, Sophie L., et al.
Published: (2026)

Advancing Academic Knowledge Retrieval via LLM-enhanced Representation Similarity Fusion
by: Dai, Wei, et al.
Published: (2024)

Unintended Impacts of LLM Alignment on Global Representation
by: Ryan, Michael J., et al.
Published: (2024)

TruthRL: Incentivizing Truthful LLMs via Reinforcement Learning
by: Wei, Zhepei, et al.
Published: (2025)

Compressing LLMs: The Truth is Rarely Pure and Never Simple
by: Jaiswal, Ajay, et al.
Published: (2023)

How Post-Training Reshapes LLMs: A Mechanistic View on Knowledge, Truthfulness, Refusal, and Confidence
by: Du, Hongzhe, et al.
Published: (2025)

LLM Flow Processes for Text-Conditioned Regression
by: Biggs, Felix, et al.
Published: (2026)

Bridge: A Unified Framework to Knowledge Graph Completion via Language Models and Knowledge Representation
by: Qiao, Qiao, et al.
Published: (2024)

Enhancing LLM Knowledge Learning through Generalization
by: Zhu, Mingkang, et al.
Published: (2025)

TruthX: Alleviating Hallucinations by Editing Large Language Models in Truthful Space
by: Zhang, Shaolei, et al.
Published: (2024)

Multimodal Contrastive Representation Learning in Augmented Biomedical Knowledge Graphs
by: Dang, Tien, et al.
Published: (2025)

Representation Consistency for Accurate and Coherent LLM Answer Aggregation
by: Jiang, Junqi, et al.
Published: (2025)

Advancing LLM Safe Alignment with Safety Representation Ranking
by: Du, Tianqi, et al.
Published: (2025)

Improving LLM Final Representations with Inter-Layer Geometry
by: Ulanovski, Tom, et al.
Published: (2026)

Understanding the Detrimental Class-level Effects of Data Augmentation
by: Kirichenko, Polina, et al.
Published: (2023)

Superficial Safety Alignment Hypothesis
by: Li, Jianwei, et al.
Published: (2024)

Mitigating Heterogeneous Token Overfitting in LLM Knowledge Editing
by: Liu, Tianci, et al.
Published: (2025)

Controlled LLM Decoding via Discrete Auto-regressive Biasing
by: Pynadath, Patrick, et al.
Published: (2025)

Layerwise Recall and the Geometry of Interwoven Knowledge in LLMs
by: Lei, Ge, et al.
Published: (2025)