When Do Models Admit They Are Wrong? Failure Disclosure Is Unstable Under Reinforcement Learning
cs.LG, cs.AI, cs.CL
Submitted: 2026-09-27
Updated: 2026-09-27
Code: https://github.com/safety-research/failure-disclosure
Terminology
Sources
- Phi-4 Technical Report
- Program Synthesis with Large Language Models
- How LLM Task-Adaptation Reshapes Alignment: A Multi-dimensional Study of Behavioral and Representational Drift
- When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors
- Gemma 2: Improving Open Language Models at a Practical Size
- LLM-42: Enabling Determinism in LLM Inference with Verified Speculation
- The Llama 3 Herd of Models
- Behavior-Consistent Deep Reinforcement Learning
- Training LLMs for Honesty via Confessions
- Language Models (Mostly) Know What They Know
- HalluWorld: A Controlled Benchmark for Hallucination via Reference World Models
- Reasoning Under Pressure: How do Training Incentives Influence Chain-of-Thought Monitorability?
- Self-Evaluating LLMs for Multi-Step Tasks: Stepwise Confidence Estimation for Failure Detection
- Qwen2.5 Technical Report
- Overtrained, Not Misaligned
- Proximal Policy Optimization Algorithms
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- 2 OLMo 2 Furious
- Know When You're Wrong: Aligning Confidence with Correctness for LLM Error Detection
- Can Aha Moments Be Fake? Towards Quantifying Decorative and True Thinking in Chain-of-Thought
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: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks