Copertina

How Post-Training Reshapes LLMs: A Mechanistic View on Knowledge, Truthfulness, Refusal, and Confidence

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Du, Hongzhe, Li, Weikai, Cai, Min, Saraipour, Karim, Zhang, Zimin, Lakkaraju, Himabindu, Sun, Yizhou, Zhang, Shichang
Natura:	Preprint
Pubblicazione:	2025
Soggetti:	Computation and Language Artificial Intelligence Machine Learning
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

From Indirect Object Identification to Syllogisms: Exploring Binary Mechanisms in Transformer Circuits
di: Saraipour, Karim, et al.
Pubblicazione: (2025)

Who Gets Credit or Blame? Attributing Accountability in Modern AI Systems
di: Zhang, Shichang, et al.
Pubblicazione: (2025)

Towards Unified Attribution in Explainable AI, Data-Centric AI, and Mechanistic Interpretability
di: Zhang, Shichang, et al.
Pubblicazione: (2025)

Understanding the Effects of Iterative Prompting on Truthfulness
di: Krishna, Satyapriya, et al.
Pubblicazione: (2024)

Generalized Group Data Attribution
di: Ley, Dan, et al.
Pubblicazione: (2024)

Discriminative Feature Attributions: Bridging Post Hoc Explainability and Inherent Interpretability
di: Bhalla, Usha, et al.
Pubblicazione: (2023)

Monitorability as a Free Gift: How RLVR Spontaneously Aligns Reasoning
di: Xiong, Zidi, et al.
Pubblicazione: (2026)

Explaining the Model, Protecting Your Data: Revealing and Mitigating the Data Privacy Risks of Post-Hoc Model Explanations via Membership Inference
di: Huang, Catherine, et al.
Pubblicazione: (2024)

Learning Recourse Costs from Pairwise Feature Comparisons
di: Rawal, Kaivalya, et al.
Pubblicazione: (2024)

Manipulating Large Language Models to Increase Product Visibility
di: Kumar, Aounon, et al.
Pubblicazione: (2024)

Towards Uncovering How Large Language Model Works: An Explainability Perspective
di: Zhao, Haiyan, et al.
Pubblicazione: (2024)

Automated Molecular Concept Generation and Labeling with Large Language Models
di: Zhang, Zimin, et al.
Pubblicazione: (2024)

Confronting LLMs with Traditional ML: Rethinking the Fairness of Large Language Models in Tabular Classifications
di: Liu, Yanchen, et al.
Pubblicazione: (2023)

In-Context Unlearning: Language Models as Few Shot Unlearners
di: Pawelczyk, Martin, et al.
Pubblicazione: (2023)

Characterizing Data Point Vulnerability via Average-Case Robustness
di: Han, Tessa, et al.
Pubblicazione: (2023)

On the Impact of Fine-Tuning on Chain-of-Thought Reasoning
di: Lobo, Elita, et al.
Pubblicazione: (2024)

On the Trade-offs between Adversarial Robustness and Actionable Explanations
di: Krishna, Satyapriya, et al.
Pubblicazione: (2023)

Computational Copyright: Towards A Royalty Model for Music Generative AI
di: Deng, Junwei, et al.
Pubblicazione: (2023)

In-Context Explainers: Harnessing LLMs for Explaining Black Box Models
di: Kroeger, Nicholas, et al.
Pubblicazione: (2023)

More RLHF, More Trust? On The Impact of Preference Alignment On Trustworthiness
di: Li, Aaron J., et al.
Pubblicazione: (2024)

Faithfulness vs. Plausibility: On the (Un)Reliability of Explanations from Large Language Models
di: Agarwal, Chirag, et al.
Pubblicazione: (2024)

Over-Refusal and Representation Subspaces: A Mechanistic Analysis of Task-Conditioned Refusal in Aligned LLMs
di: Maskey, Utsav, et al.
Pubblicazione: (2026)

Follow My Instruction and Spill the Beans: Scalable Data Extraction from Retrieval-Augmented Generation Systems
di: Qi, Zhenting, et al.
Pubblicazione: (2024)

Measuring the Faithfulness of Thinking Drafts in Large Reasoning Models
di: Xiong, Zidi, et al.
Pubblicazione: (2025)

Towards Unifying Interpretability and Control: Evaluation via Intervention
di: Bhalla, Usha, et al.
Pubblicazione: (2024)

Interpretability Needs a New Paradigm
di: Madsen, Andreas, et al.
Pubblicazione: (2024)

MedSafetyBench: Evaluating and Improving the Medical Safety of Large Language Models
di: Han, Tessa, et al.
Pubblicazione: (2024)

Operationalizing the Blueprint for an AI Bill of Rights: Recommendations for Practitioners, Researchers, and Policy Makers
di: Oesterling, Alex, et al.
Pubblicazione: (2024)

Beyond Over-Refusal: Scenario-Based Diagnostics and Post-Hoc Mitigation for Exaggerated Refusals in LLMs
di: Yuan, Shuzhou, et al.
Pubblicazione: (2025)

EvoLM: In Search of Lost Language Model Training Dynamics
di: Qi, Zhenting, et al.
Pubblicazione: (2025)

Generalizing Trust: Weak-to-Strong Trustworthiness in Language Models
di: Pawelczyk, Martin, et al.
Pubblicazione: (2024)

Mechanistic Insights into Loop Dynamics in Reshaping the Conformation of YB‐1 Cold Shock Domain
di: Min Sun, et al.
Pubblicazione: (2025)

The Struggle Between Continuation and Refusal: A Mechanistic Analysis of the Continuation-Triggered Jailbreak in LLMs
di: Deng, Yonghong, et al.
Pubblicazione: (2026)

Adapting Fine-Grained Cross-View Localization to Areas without Fine Ground Truth
di: Xia, Zimin, et al.
Pubblicazione: (2024)

On the Hardness of Faithful Chain-of-Thought Reasoning in Large Language Models
di: Tanneru, Sree Harsha, et al.
Pubblicazione: (2024)

Evaluating Adversarial Robustness of Concept Representations in Sparse Autoencoders
di: Li, Aaron J., et al.
Pubblicazione: (2025)

Detecting LLM-Generated Peer Reviews
di: Rao, Vishisht, et al.
Pubblicazione: (2025)

Adapting to LLMs: How Insiders and Outsiders Reshape Scientific Knowledge Production
di: Xu, Huimin, et al.
Pubblicazione: (2025)

Rejection Improves Reliability: Training LLMs to Refuse Unknown Questions Using RL from Knowledge Feedback
di: Xu, Hongshen, et al.
Pubblicazione: (2024)

Wired for Overconfidence: A Mechanistic Perspective on Inflated Verbalized Confidence in LLMs
di: Zhao, Tianyi, et al.
Pubblicazione: (2026)