Reward Bias Substitution: Single-Axis Bias Mitigations Redirect Optimization Pressure
cs.AI
Submitted: 2026-05-27
Updated: 2026-11-07
Code: https://github.com/maxlampe/bias_substitution
Terminology
Sources
- Training Verifiers to Solve Math Word Problems
- One Bias After Another: Mechanistic Reward Shaping and Persistent Biases in Language Reward Models
- C2PO: Diagnosing and Disentangling Bias Shortcuts in LLMs
- The Llama 3 Herd of Models
- Measuring and mitigating overreliance to build human-compatible AI
- Mitigating Length Bias in RLHF through a Causal Lens
- Length-Controlled Margin-Based Preference Optimization without Reference Model
- Eliminating Inductive Bias in Reward Models with Information-Theoretic Guidance
- Skywork-Reward-V2: Scaling Preference Data Curation via Human-AI Synergy
- Debiasing Reward Models by Representation Learning with Guarantees
- Nemotron-4 340B Technical Report
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- How RLHF Amplifies Sycophancy
- General Preference Reinforcement Learning
- Beyond Reward Hacking: Causal Rewards for Large Language Model Alignment
- Shorten After You're Right: Lazy Length Penalties for Reasoning RL
- Bias Fitting to Mitigate Length Bias of Reward Model in RLHF
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