Dialectics of Alignment: Harnessing Unsafe Knowledge for Dynamic Safety Routing
cs.LG
Submitted: 2026-05-30
Updated: 2026-10-02
Comments: Accepted to The Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS) 2026. First two authors contributed equally
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
- Red-Teaming Large Language Models using Chain of Utterances for Safety-Alignment
- OR-Bench: An Over-Refusal Benchmark for Large Language Models
- DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
- DeepSeek-V3 Technical Report
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- Oyster-I: Beyond Refusal -- Constructive Safety Alignment for Responsible Language Models
- Mistral 7B
- Mixtral of Experts
- On Evaluating LLM Alignment by Evaluating LLMs as Judges
- Llama 2: Open Foundation and Fine-Tuned Chat Models
- Zephyr: Direct Distillation of LM Alignment
- Qwen2.5 Technical Report
- RSafe: Incentivizing proactive reasoning to build robust and adaptive LLM safeguards
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks