Training Language Models to Fail Gracefully

cs.LG

Submitted: 2026-10-14

Updated: 2026-10-14

Comments: preprint

Code: https://github.com/radixark/miles

Project page: https://zanette-labs.github.io/failing-gracefully

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Sources

• Beyond Binary Rewards: Training LMs to Reason About Their Uncertainty
• Why Language Models Hallucinate
• KARL: Knowledge-Aware Reasoning and Reinforcement Learning for Knowledge-Intensive Visual Grounding
• Delightful Gradients Accelerate Corner Escape
• OpenAI o1 System Card
• Delightful Policy Gradient
• TIAR: Trajectory-Informed Advantage Reweighting for LLM Abstention Learning
• Tail-Likelihood Reinforcement Learning
• Proximal Policy Optimization Algorithms
• HybridFlow: A Flexible and Efficient RLHF Framework
• Recovery RL: Safe Reinforcement Learning with Learned Recovery Zones
• Rejection Improves Reliability: Training LLMs to Refuse Unknown Questions Using RL from Knowledge Feedback
• Qwen3 Technical Report
• DAPO: An Open-Source LLM Reinforcement Learning System at Scale
• GRACE: Reinforcement Learning for Grounded Response and Abstention under Contextual Evidence

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: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks