Credence Calibration Game? Calibrating Large Language Models through Structured Play

2025-08-19

AI Summary

Large language models (LLMs) often exhibit miscalibrated confidence—i.e., their predicted confidence scores poorly align with actual answer correctness—posing risks in high-stakes decision-making. Method: We propose “Credibility Calibration Game,” a training-free, fine-tuning-free dynamic calibration framework. It employs a structured, game-inspired prompting mechanism to enable iterative human–model interaction: the model generates an answer with confidence, receives natural-language feedback, synthesizes a summary, and then performs more deliberate reasoning and self-calibration in subsequent rounds. Contribution/Results: This work is the first to introduce gamified interaction for LLM confidence calibration under fully unsupervised, parameter-frozen conditions. Experiments across diverse open- and closed-source models (e.g., Llama-3, GPT-4) and tasks (factual QA, mathematical reasoning) show significant improvements in calibration metrics—including Expected Calibration Error (ECE) and Brier Score—with strong generalization and plug-and-play applicability.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Cognitive Modeling & Cognitive Systems: Computational Creativity
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

As Large Language Models (LLMs) are increasingly deployed in decision-critical domains, it becomes essential to ensure that their confidence estimates faithfully correspond to their actual correctness. Existing calibration methods have primarily focused on post-hoc adjustments or auxiliary model training; however, many of these approaches necessitate additional supervision or parameter updates. In this work, we propose a novel prompt-based calibration framework inspired by the Credence Calibration Game. Our method establishes a structured interaction loop wherein LLMs receive feedback based on the alignment of their predicted confidence with correctness. Through feedback-driven prompting and natural language summaries of prior performance, our framework dynamically improves model calibration. Extensive experiments across models and game configurations demonstrate consistent improvements in evaluation metrics. Our results highlight the potential of game-based prompting as an effective strategy for LLM calibration. Code and data are available at https://anonymous.4open.science/r/LLM-Calibration/.

Problem

• Calibrating LLM confidence to match actual correctness
• Reducing need for supervision or parameter updates
• Improving calibration through structured feedback interaction

Innovation

• Structured feedback interaction loop
• Feedback-driven prompting for calibration
• Natural language performance summaries