Towards Multi-Value Alignment via Value Localization in LLMs
cs.AI
Submitted: 2026-11-02
Updated: 2026-11-02
Code: https://github.com/sihuo-design/ValueHeads
Terminology
Sources
- GPT-4 Technical Report
- Understanding intermediate layers using linear classifier probes
- Deep Reinforcement Learning from Policy-Dependent Human Feedback
- A General Language Assistant as a Laboratory for Alignment
- Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
- What Does BERT Look At? An Analysis of BERT's Attention
- Contextual Moral Value Alignment Through Context-Based Aggregation
- Aligned Multi Objective Optimization
- The Llama 3 Herd of Models
- Catastrophic Jailbreak of Open-source LLMs via Exploiting Generation
- AI Alignment: A Comprehensive Survey
- Investigating The Functional Roles of Attention Heads in Vision Language Models: Evidence for Reasoning Modules
- OrthAlign: Orthogonal Subspace Decomposition for Non-Interfering Multi-Objective Alignment
- Cognitive Mirrors: Exploring the Diverse Functional Roles of Attention Heads in LLM Reasoning
- Pointer Sentinel Mixture Models
- In-context Learning and Induction Heads
- Layer by Layer: Uncovering Hidden Representations in Language Models
- BERT Rediscovers the Classical NLP Pipeline
- LLaMA: Open and Efficient Foundation Language Models
- Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Pruned
Related papers
- Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Embeddings for Preferences, Not Semantics
- Can We Trust Item Response Theory for AI Evaluation?
- Diagnosing Task Insensitivity in Language Agents