Researchers found that calibrated vision-language models can still report high confidence in incorrect answers due to trajectory-independent verbalized confidence. They propose the Trajectory-Grounding Score (TGS) to address this issue, which evaluates a model's confidence based on its reasoning trajectory. This discovery highlights a blind spot in current evaluation practices for AI models.

Researchers propose fine-grained confidence calibration methods for Large Language Models (LLMs) in automated code revision tasks, improving calibration error and enabling more trustworthy model usage.

Researchers introduced Reinforcement Learning from Self-Feedback (RLSF), a post-training stage that refines large language models by leveraging their intrinsic confidence as a self-generated reward. This approach refines probability estimates and strengthens step-by-step reasoning, improving performance on arithmetic reasoning and multiple-choice question answering.