But what is your honest answer? Aiding LLM-judges with honest alternatives using steering vectors
cs.LG, cs.AI
Submitted: 2025-05-23
Updated: 2026-11-16
Code: https://github.com/watermeleon/judge_with_steered_response
Terminology
Sources
- Comparing Bottom-Up and Top-Down Steering Approaches on In-Context Learning Tasks
- Discovering Latent Knowledge in Language Models Without Supervision
- Improving Steering Vectors by Targeting Sparse Autoencoder Features
- White-Box Sensitivity Auditing with Steering Vectors
- Alignment faking in large language models
- Frontier Models are Capable of In-context Scheming
- Strategic Dishonesty Can Undermine AI Safety Evaluations of Frontier LLMs
- Steering Llama 2 via Contrastive Activation Addition
- Towards Understanding Sycophancy in Language Models
- Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis