Image de couverture de livre

Extracting alignment data in open models

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Barbero, Federico, Gu, Xiangming, Choquette-Choo, Christopher A., Sitawarin, Chawin, Jagielski, Matthew, Yona, Itay, Veličković, Petar, Shumailov, Ilia, Hayes, Jamie
Format:	Preprint
Publié:	2025
Sujets:	Artificial Intelligence
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Interpreting the Repeated Token Phenomenon in Large Language Models
par: Yona, Itay, et autres
Publié: (2025)

Stealing User Prompts from Mixture of Experts
par: Yona, Itay, et autres
Publié: (2024)

Buffer Overflow in Mixture of Experts
par: Hayes, Jamie, et autres
Publié: (2024)

Soft Instruction De-escalation Defense
par: Walter, Nils Philipp, et autres
Publié: (2025)

UnUnlearning: Unlearning is not sufficient for content regulation in advanced generative AI
par: Shumailov, Ilia, et autres
Publié: (2024)

Measuring memorization in language models via probabilistic extraction
par: Hayes, Jamie, et autres
Publié: (2024)

Lessons from Defending Gemini Against Indirect Prompt Injections
par: Shi, Chongyang, et autres
Publié: (2025)

Beyond Slow Signs in High-fidelity Model Extraction
par: Foerster, Hanna, et autres
Publié: (2024)

Cascading Adversarial Bias from Injection to Distillation in Language Models
par: Chaudhari, Harsh, et autres
Publié: (2025)

Softmax is not Enough (for Sharp Size Generalisation)
par: Veličković, Petar, et autres
Publié: (2024)

Are aligned neural networks adversarially aligned?
par: Carlini, Nicholas, et autres
Publié: (2023)

Exploring the limits of strong membership inference attacks on large language models
par: Hayes, Jamie, et autres
Publié: (2025)

Perplexity Cannot Always Tell Right from Wrong
par: Veličković, Petar, et autres
Publié: (2026)

Positional Embedding-Aware Activations
par: Shah, Kathan, et autres
Publié: (2023)

PAL: Proxy-Guided Black-Box Attack on Large Language Models
par: Sitawarin, Chawin, et autres
Publié: (2024)

Why do LLMs attend to the first token?
par: Barbero, Federico, et autres
Publié: (2025)

How do LLMs Compute Verbal Confidence
par: Kumaran, Dharshan, et autres
Publié: (2026)

LLMs unlock new paths to monetizing exploits
par: Carlini, Nicholas, et autres
Publié: (2025)

Auditing Private Prediction
par: Chadha, Karan, et autres
Publié: (2024)

Thought-Transfer: Indirect Targeted Poisoning Attacks on Chain-of-Thought Reasoning Models
par: Chaudhari, Harsh, et autres
Publié: (2026)

Language Models May Verbatim Complete Text They Were Not Explicitly Trained On
par: Liu, Ken Ziyu, et autres
Publié: (2025)

Mark My Words: Analyzing and Evaluating Language Model Watermarks
par: Piet, Julien, et autres
Publié: (2023)

Recurrent Aggregators in Neural Algorithmic Reasoning
par: Xu, Kaijia, et autres
Publié: (2024)

PubDef: Defending Against Transfer Attacks From Public Models
par: Sitawarin, Chawin, et autres
Publié: (2023)

Measuring memorization in RLHF for code completion
par: Pappu, Aneesh, et autres
Publié: (2024)

Reasoning Introduces New Poisoning Attacks Yet Makes Them More Complicated
par: Foerster, Hanna, et autres
Publié: (2025)

Does More Inference-Time Compute Really Help Robustness?
par: Wu, Tong, et autres
Publié: (2025)

Locking Machine Learning Models into Hardware
par: Clifford, Eleanor, et autres
Publié: (2024)

Machine Learning needs Better Randomness Standards: Randomised Smoothing and PRNG-based attacks
par: Dahiya, Pranav, et autres
Publié: (2023)

Understanding Performance Gap Between Parallel and Sequential Sampling in Large Reasoning Models
par: Gu, Xiangming, et autres
Publié: (2026)

Inexact Unlearning Needs More Careful Evaluations to Avoid a False Sense of Privacy
par: Hayes, Jamie, et autres
Publié: (2024)

Operationalizing Contextual Integrity in Privacy-Conscious Assistants
par: Ghalebikesabi, Sahra, et autres
Publié: (2024)

Cayley Graph Propagation
par: Wilson, JJ, et autres
Publié: (2024)

Friends and Grandmothers in Silico: Localizing Entity Cells in Language Models
par: Yona, Itay, et autres
Publié: (2026)

Honeyval: A Comprehensive Evaluation Framework for LLM-powered HTTP Honeypots
par: Vero, Mark, et autres
Publié: (2026)

Round and Round We Go! What makes Rotary Positional Encodings useful?
par: Barbero, Federico, et autres
Publié: (2024)

The Last Iterate Advantage: Empirical Auditing and Principled Heuristic Analysis of Differentially Private SGD
par: Steinke, Thomas, et autres
Publié: (2024)

Privacy Ripple Effects from Adding or Removing Personal Information in Language Model Training
par: Borkar, Jaydeep, et autres
Publié: (2025)

OODRobustBench: a Benchmark and Large-Scale Analysis of Adversarial Robustness under Distribution Shift
par: Li, Lin, et autres
Publié: (2023)

StruQ: Defending Against Prompt Injection with Structured Queries
par: Chen, Sizhe, et autres
Publié: (2024)