Copertina

Rethinking Safety in LLM Fine-tuning: An Optimization Perspective

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Kim, Minseon, Kwak, Jin Myung, Alssum, Lama, Ghanem, Bernard, Torr, Philip, Krueger, David, Barez, Fazl, Bibi, Adel
Natura:	Preprint
Pubblicazione:	2025
Soggetti:	Machine Learning Artificial Intelligence
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

Unforgotten Safety: Preserving Safety Alignment of Large Language Models with Continual Learning
di: Alssum, Lama, et al.
Pubblicazione: (2025)

Beyond Linear Probes: Dynamic Safety Monitoring for Language Models
di: Oldfield, James, et al.
Pubblicazione: (2025)

Towards Interpretable Sequence Continuation: Analyzing Shared Circuits in Large Language Models
di: Lan, Michael, et al.
Pubblicazione: (2023)

Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencoders
di: Lan, Michael, et al.
Pubblicazione: (2024)

PoisonBench: Assessing Large Language Model Vulnerability to Data Poisoning
di: Fu, Tingchen, et al.
Pubblicazione: (2024)

Understanding Addition in Transformers
di: Quirke, Philip, et al.
Pubblicazione: (2023)

SafetyNet: Detecting Harmful Outputs in LLMs by Modeling and Monitoring Deceptive Behaviors
di: Chaudhary, Maheep, et al.
Pubblicazione: (2025)

Rethinking AI Cultural Alignment
di: Bravansky, Michal, et al.
Pubblicazione: (2025)

Bi-Factorial Preference Optimization: Balancing Safety-Helpfulness in Language Models
di: Zhang, Wenxuan, et al.
Pubblicazione: (2024)

Model Merging and Safety Alignment: One Bad Model Spoils the Bunch
di: Hammoud, Hasan Abed Al Kader, et al.
Pubblicazione: (2024)

Same Question, Different Words: A Latent Adversarial Framework for Prompt Robustness
di: Fu, Tingchen, et al.
Pubblicazione: (2025)

Towards Interpreting Visual Information Processing in Vision-Language Models
di: Neo, Clement, et al.
Pubblicazione: (2024)

Query Circuits: Explaining How Language Models Answer User Prompts
di: Wu, Tung-Yu, et al.
Pubblicazione: (2025)

On Pretraining Data Diversity for Self-Supervised Learning
di: Hammoud, Hasan Abed Al Kader, et al.
Pubblicazione: (2024)

SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training?
di: Hammoud, Hasan Abed Al Kader, et al.
Pubblicazione: (2024)

Prompting a Pretrained Transformer Can Be a Universal Approximator
di: Petrov, Aleksandar, et al.
Pubblicazione: (2024)

Do Sparse Autoencoders Generalize? A Case Study of Answerability
di: Heindrich, Lovis, et al.
Pubblicazione: (2025)

Purifying Task Vectors in Knowledge-Aware Subspace for Model Merging
di: An, Bang, et al.
Pubblicazione: (2025)

Open Problems in Machine Unlearning for AI Safety
di: Barez, Fazl, et al.
Pubblicazione: (2025)

FedMedICL: Towards Holistic Evaluation of Distribution Shifts in Federated Medical Imaging
di: Alhamoud, Kumail, et al.
Pubblicazione: (2024)

Align to Misalign: Automatic LLM Jailbreak with Meta-Optimized LLM Judges
di: Koo, Hamin, et al.
Pubblicazione: (2025)

Interpreting Learned Feedback Patterns in Large Language Models
di: Marks, Luke, et al.
Pubblicazione: (2023)

Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoders
di: Marks, Luke, et al.
Pubblicazione: (2024)

Large Language Models Relearn Removed Concepts
di: Lo, Michelle, et al.
Pubblicazione: (2024)

A Guardrail for Safety Preservation: When Safety-Sensitive Subspace Meets Harmful-Resistant Null-Space
di: Zhang, Bingjie, et al.
Pubblicazione: (2025)

VAL-Bench: Belief Consistency as a measure for Value Alignment in Language Models
di: Gupta, Aman, et al.
Pubblicazione: (2025)

Shh, don't say that! Domain Certification in LLMs
di: Emde, Cornelius, et al.
Pubblicazione: (2025)

Interpreting Context Look-ups in Transformers: Investigating Attention-MLP Interactions
di: Neo, Clement, et al.
Pubblicazione: (2024)

Do as I do (Safely): Mitigating Task-Specific Fine-tuning Risks in Large Language Models
di: Eiras, Francisco, et al.
Pubblicazione: (2024)

Continual Learning on a Diet: Learning from Sparsely Labeled Streams Under Constrained Computation
di: Zhang, Wenxuan, et al.
Pubblicazione: (2024)

Towards Understanding Subliminal Learning: When and How Hidden Biases Transfer
di: Schrodi, Simon, et al.
Pubblicazione: (2025)

Visualizing Neural Network Imagination
di: Wichers, Nevan, et al.
Pubblicazione: (2024)

Towards Interpretable Deep Local Learning with Successive Gradient Reconciliation
di: Yang, Yibo, et al.
Pubblicazione: (2024)

On the Coexistence and Ensembling of Watermarks
di: Petrov, Aleksandar, et al.
Pubblicazione: (2025)

CyberLLMInstruct: A Pseudo-malicious Dataset Revealing Safety-performance Trade-offs in Cyber Security LLM Fine-tuning
di: ElZemity, Adel, et al.
Pubblicazione: (2025)

Understanding Addition and Subtraction in Transformers
di: Quirke, Philip, et al.
Pubblicazione: (2024)

Universal In-Context Approximation By Prompting Fully Recurrent Models
di: Petrov, Aleksandar, et al.
Pubblicazione: (2024)

Chain-of-Thought Hijacking
di: Zhao, Jianli, et al.
Pubblicazione: (2025)

ToolTweak: An Attack on Tool Selection in LLM-based Agents
di: Sneh, Jonathan, et al.
Pubblicazione: (2025)

OMNI-LEAK: Orchestrator Multi-Agent Network Induced Data Leakage
di: Naik, Akshat, et al.
Pubblicazione: (2026)