Capability Self-Assessment: Teaching LLMs to Know Their Limits

2026-05-29

AI Summary

This work addresses the tendency of large language models to overestimate their capabilities and struggle with accurately assessing task solvability. It introduces, for the first time, capability self-assessment (CSA) as a policy learning problem, training models via reinforcement learning to dynamically decide whether to answer a query themselves or delegate it elsewhere. This approach preserves the model’s original competencies while significantly improving the accuracy of self-assessment. Compared to supervised fine-tuning, reinforcement learning yields markedly better CSA performance and demonstrates strong out-of-distribution generalization. The proposed mechanism effectively enhances decision-making in local–cloud collaborative inference and optimizes the selection of training data.

Abstract

The ability to recognize one's own limitations and decide whether to solve a problem or delegate is fundamental for reliable intelligent systems. Yet we show that modern large language models systematically lack this ability: across diverse model families and scales, they overestimate their competence and attempt queries they cannot solve. We refer to this ability as Capability Self-Assessment (CSA) and formulate it as a policy-learning problem, aiming to improve self-assessment while preserving the model's original capabilities. Our results show that reinforcement learning teaches CSA effectively, significantly outperforming supervised fine-tuning while preserving original capabilities. In contrast, supervised fine-tuning severely degrades the capabilities the model is meant to assess. Moreover, learned self-assessment behavior generalizes well out of distribution, suggesting that CSA is a transferable model trait. Finally, CSA is practically useful: it improves local-cloud decision making at inference time and provides a signal for targeted data selection during training.

Problem

• Capability Self-Assessment
• Large Language Models
• Self-Awareness
• Overconfidence
• Delegation

Innovation

• Capability Self-Assessment
• Reinforcement Learning
• Out-of-Distribution Generalization
• Local-Cloud Decision Making
• Policy Learning