Honesty Is a Race: Predicting When Verifiable-Reward RL Teaches a Model to Bluff
cs.LG, cs.AI, cs.CL
Submitted: 2026-11-11
Updated: 2026-11-11
Terminology
Sources
- Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
- Learning to Give Checkable Answers with Prover-Verifier Games
- Training Language Models to Fail Gracefully
- Program Synthesis with Large Language Models
- The Internal State of an LLM Knows When It's Lying
- Prompt Interference in LLM Post-Training: Why Pass@k Optimization Can Degrade Pass@1
- GRPO, Dr. GRPO, and DAPO Are Three Operations on One Number: The Group-Standard-Deviation Identity
- Uncalibrated Reasoning: GRPO Induces Overconfidence for Stochastic Outcomes
- Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations
- Reward Hacking in Language Model Agents: Revisiting AI Safety Gridworlds
- Reinforcement Learning with Verifiable yet Noisy Rewards under Imperfect Verifiers
- Localizing Lying in Llama: Understanding Instructed Dishonesty on True-False Questions Through Prompting, Probing, and Patching
- Abstention as an Action Can Kill Both the Reward Gradient and the KL Anchor: Collapse Law and Repair for Error-Penalized Reinforcement Learning
- Rethinking GRPO for Large Reasoning Model Safety Alignment
- The Unlearnability Phenomenon in RLVR for Language Models
- Euston: Training Away Mathematical Sycophancy Without Losing the Mathematics
- Beyond Binary Rewards: Training LMs to Reason About Their Uncertainty
- Beyond the Verdict: Agentic Code Judging with Verifiable Feedback
- Collective Dynamics of Self-Reinforcement: How Cross-Task Coupling Drives Policy Collapse in Unsupervised RLVR
- Delay, Plateau, or Collapse: Evaluating the Impact of Systematic Verification Error on RLVR
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