Imagen de Portada

SALT: Steering Activations towards Leakage-free Thinking in Chain of Thought

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Batra, Shourya, Tillman, Pierce, Gaggar, Samarth, Kesineni, Shashank, Zhu, Kevin, Dev, Sunishchal, Panda, Ashwinee, Sharma, Vasu, Chaudhary, Maheep
Formato:	Preprint
Publicado:	2025
Materias:	Cryptography and Security Artificial Intelligence Computation and Language Machine Learning
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Amortized Latent Steering: Low-Cost Alternative to Test-Time Optimization
por: Egbuna, Nathan, et al.
Publicado: (2025)

FRIT: Using Causal Importance to Improve Chain-of-Thought Faithfulness
por: Swaroop, Anand, et al.
Publicado: (2025)

Peek-a-Boo Reasoning: Contrastive Region Masking in MLLMs
por: Chaturvedi, Isha, et al.
Publicado: (2025)

Evaluation Awareness Scales Predictably in Open-Weights Large Language Models
por: Chaudhary, Maheep, et al.
Publicado: (2025)

Broken Chains: The Cost of Incomplete Reasoning in LLMs
por: Su, Ian, et al.
Publicado: (2026)

Alignment-Constrained Dynamic Pruning for LLMs: Identifying and Preserving Alignment-Critical Circuits
por: Patel, Dev, et al.
Publicado: (2025)

Hydra: A Modular Architecture for Efficient Long-Context Reasoning
por: Chaudhary, Siddharth, et al.
Publicado: (2025)

In-Context Environments Induce Evaluation-Awareness in Language Models
por: Chaudhary, Maheep
Publicado: (2026)

PALADIN: Self-Correcting Language Model Agents to Cure Tool-Failure Cases
por: Vuddanti, Sri Vatsa, et al.
Publicado: (2025)

Modeling and Predicting Multi-Turn Answer Instability in Large Language Models
por: He, Jiahang, et al.
Publicado: (2025)

Activation Steering for Chain-of-Thought Compression
por: Azizi, Seyedarmin, et al.
Publicado: (2025)

Inference-Time Chain-of-Thought Pruning with Latent Informativeness Signals
por: Li, Sophie, et al.
Publicado: (2025)

Optimizing Chain-of-Thought Confidence via Topological and Dirichlet Risk Analysis
por: More, Abhishek, et al.
Publicado: (2025)

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

Evaluating Open-Source Sparse Autoencoders on Disentangling Factual Knowledge in GPT-2 Small
por: Chaudhary, Maheep, et al.
Publicado: (2024)

DuoLens: A Framework for Robust Detection of Machine-Generated Multilingual Text and Code
por: Agrawal, Shriyansh, et al.
Publicado: (2025)

ProMoral-Bench: Evaluating Prompting Strategies for Moral Reasoning and Safety in LLMs
por: Thomas, Rohan Subramanian, et al.
Publicado: (2026)

Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT?
por: Nunez, Jeanmely Rojas, et al.
Publicado: (2026)

A Few Bad Neurons: Isolating and Surgically Correcting Sycophancy
por: O'Brien, Claire, et al.
Publicado: (2026)

Is Chain-of-Thought Really Not Explainability? Chain-of-Thought Can Be Faithful without Hint Verbalization
por: Zaman, Kerem, et al.
Publicado: (2025)

From Refusal Tokens to Refusal Control: Discovering and Steering Category-Specific Refusal Directions
por: Alagharu, Rishab, et al.
Publicado: (2026)

COMPASS: Context-Modulated PID Attention Steering System for Hallucination Mitigation
por: Sahay, Kenji, et al.
Publicado: (2025)

AgentChangeBench: A Multi-Dimensional Evaluation Framework for Goal-Shift Robustness in Conversational AI
por: Rana, Manik, et al.
Publicado: (2025)

@GrokSet: multi-party Human-LLM Interactions in Social Media
por: Migliarini, Matteo, et al.
Publicado: (2026)

Freespace twistronics for optical supertopologies
por: Dev, Vasu, et al.
Publicado: (2025)

Emergent Misalignment via In-Context Learning: Narrow in-context examples can produce broadly misaligned LLMs
por: Afonin, Nikita, et al.
Publicado: (2025)

Decoding Answers Before Chain-of-Thought: Evidence from Pre-CoT Probes and Activation Steering
por: Cox, Kyle, et al.
Publicado: (2026)

Data Augmentation for NeRFs in the Low Data Limit
por: Gaggar, Ayush, et al.
Publicado: (2025)

GeoSteer: Faithful Chain-of-Thought Steering via Latent Manifold Gradients
por: Kazama, Kentaro, et al.
Publicado: (2026)

Evaluating Chain-of-Thought Reasoning through Reusability and Verifiability
por: Aggarwal, Shashank, et al.
Publicado: (2026)

Melody or Machine: Detecting Synthetic Music with Dual-Stream Contrastive Learning
por: Batra, Arnesh, et al.
Publicado: (2025)

Deep Thinking by Markov Chain of Continuous Thoughts
por: Liu, Jiayu, et al.
Publicado: (2025)

Safer Reasoning Traces: Measuring and Mitigating Chain-of-Thought Leakage in LLMs
por: Ahrend, Patrick, et al.
Publicado: (2026)

PII Jailbreaking in LLMs via Activation Steering Reveals Personal Information Leakage
por: Nakka, Krishna Kanth, et al.
Publicado: (2025)

How does Chain of Thought Think? Mechanistic Interpretability of Chain-of-Thought Reasoning with Sparse Autoencoding
por: Chen, Xi, et al.
Publicado: (2025)

LoRI: Reducing Cross-Task Interference in Multi-Task Low-Rank Adaptation
por: Zhang, Juzheng, et al.
Publicado: (2025)

MemeCLIP: Leveraging CLIP Representations for Multimodal Meme Classification
por: Shah, Siddhant Bikram, et al.
Publicado: (2024)

Propagation of circular Airy derivative beams in complex media
por: Kumari, Anita, et al.
Publicado: (2024)

Chain-of-Sanitized-Thoughts: Plugging PII Leakage in CoT of Large Reasoning Models
por: Das, Arghyadeep, et al.
Publicado: (2026)

Chain of Thought Still Thinks Fast: APriCoT Helps with Thinking Slow
por: Moore, Kyle, et al.
Publicado: (2024)