Cover Image

Deliberative Alignment is Deep, but Uncertainty Remains: Inference time safety improvement in reasoning via attribution of unsafe behavior to base model

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Pathmanathan, Pankayaraj, Huang, Furong
Format:	Preprint
Published:	2026
Subjects:	Machine Learning Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Using Curiosity for an Even Representation of Tasks in Continual Offline Reinforcement Learning
by: Pathmanathan, Pankayaraj, et al.
Published: (2023)

AdvBDGen: Adversarially Fortified Prompt-Specific Fuzzy Backdoor Generator Against LLM Alignment
by: Pathmanathan, Pankayaraj, et al.
Published: (2024)

Reward Models Can Improve Themselves: Reward-Guided Adversarial Failure Mode Discovery for Robust Reward Modeling
by: Pathmanathan, Pankayaraj, et al.
Published: (2025)

Is poisoning a real threat to LLM alignment? Maybe more so than you think
by: Pathmanathan, Pankayaraj, et al.
Published: (2024)

Uncertainty-Aware Deep Learning Framework for Remaining Useful Life Prediction in Turbofan Engines with Learned Aleatoric Uncertainty
by: Sharma, Krishang
Published: (2025)

Immune: Improving Safety Against Jailbreaks in Multi-modal LLMs via Inference-Time Alignment
by: Ghosal, Soumya Suvra, et al.
Published: (2024)

Deliberative Alignment: Reasoning Enables Safer Language Models
by: Guan, Melody Y., et al.
Published: (2024)

Uncovering Memorization in Timeseries Imputation models: LBRM Membership Inference and its link to attribute Leakage
by: Taleb, Faiz, et al.
Published: (2026)

Nudging: Inference-time Alignment of LLMs via Guided Decoding
by: Fei, Yu, et al.
Published: (2024)

Multi-step retrieval and reasoning improves radiology question answering with large language models
by: Wind, Sebastian, et al.
Published: (2025)

CNN-LSTM Hybrid Deep Learning Model for Remaining Useful Life Estimation
by: G, Muthukumar, et al.
Published: (2024)

SAFLEX: Self-Adaptive Augmentation via Feature Label Extrapolation
by: Ding, Mucong, et al.
Published: (2024)

Unified Gradient-Based Machine Unlearning with Remain Geometry Enhancement
by: Huang, Zhehao, et al.
Published: (2024)

Training data attribution in diffusion models via mirrored unlearning and noise-consistent skew
by: Serrà, Joan, et al.
Published: (2026)

Test-time Correlation Alignment
by: You, Linjing, et al.
Published: (2025)

Towards Reliable, Uncertainty-Aware Alignment
by: Banerjee, Debangshu, et al.
Published: (2025)

LLM-FS-Agent: A Deliberative Role-based Large Language Model Architecture for Transparent Feature Selection
by: Bal-Ghaoui, Mohamed, et al.
Published: (2025)

Agential AI for Integrated Continual Learning, Deliberative Behavior, and Comprehensible Models
by: Erden, Zeki Doruk, et al.
Published: (2025)

Context information can be more important than reasoning for time series forecasting with a large language model
by: Yang, Janghoon
Published: (2025)

Boosting Sample Efficiency and Generalization in Multi-agent Reinforcement Learning via Equivariance
by: McClellan, Joshua, et al.
Published: (2024)

BoostStep: Boosting mathematical capability of Large Language Models via improved single-step reasoning
by: Zhang, Beichen, et al.
Published: (2025)

Deep Domain Adaptation for Turbofan Engine Remaining Useful Life Prediction: Methodologies, Evaluation and Future Trends
by: Wang, Yucheng, et al.
Published: (2025)

Towards Reliable Alignment: Uncertainty-aware RLHF
by: Banerjee, Debangshu, et al.
Published: (2024)

Supervised Contrastive Learning based Dual-Mixer Model for Remaining Useful Life Prediction
by: Fu, En, et al.
Published: (2024)

Replacing thinking with tool usage enables reasoning in small language models
by: Rainone, Corrado, et al.
Published: (2025)

Energy-Efficient Wireless LLM Inference via Uncertainty and Importance-Aware Speculative Decoding
by: Park, Jihoon, et al.
Published: (2025)

Spectral Greedy Coresets for Graph Neural Networks
by: Ding, Mucong, et al.
Published: (2024)

Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL
by: Liu, Xiangyu, et al.
Published: (2023)

Interactive dense pixel visualizations for time series and model attribution explanations
by: Schlegel, Udo, et al.
Published: (2024)

Learnable Chernoff Baselines for Inference-Time Alignment
by: Madhow, Sunil, et al.
Published: (2026)

Is Best-of-N the Best of Them? Coverage, Scaling, and Optimality in Inference-Time Alignment
by: Huang, Audrey, et al.
Published: (2025)

Reinforcing privacy reasoning in LLMs via normative simulacra from fiction
by: Franchi, Matt, et al.
Published: (2026)

Training Deliberative Monitors for Black-Box Scheming Detection
by: Sinha, Aditya, et al.
Published: (2026)

Variational Inference via Smoothed Particle Hydrodynamics
by: Huang, Yongchao
Published: (2024)

Fixing confirmation bias in feature attribution methods via semantic match
by: Cinà, Giovanni, et al.
Published: (2023)

Torch-Uncertainty: A Deep Learning Framework for Uncertainty Quantification
by: Lafage, Adrien, et al.
Published: (2025)

Automated Machine Learning for Remaining Useful Life Predictions
by: Zöller, Marc-André, et al.
Published: (2023)

SAIL: Self-Improving Efficient Online Alignment of Large Language Models
by: Ding, Mucong, et al.
Published: (2024)

PoisonedParrot: Subtle Data Poisoning Attacks to Elicit Copyright-Infringing Content from Large Language Models
by: Panaitescu-Liess, Michael-Andrei, et al.
Published: (2025)

Meta-Learning and Knowledge Discovery based Physics-Informed Neural Network for Remaining Useful Life Prediction
by: Wang, Yu, et al.
Published: (2025)