Reliable confidence estimation is crucial for deploying large language models (LLMs) in confidence-aware systems, as downstream decisions such as retrieval, tool use, and adaptive computation depend on accurately estimating answer reliability. Existing approaches largely treat confidence as a property of completed responses, overlooking how confidence-related information evolves throughout the answering process.
This study investigates confidence from a temporal perspective by comparing pre-solution Feeling-of-Knowing (FOK) and post-solution Judgement-of-Learning (JOL) confidence estimates across frontier and open-source LLMs.
The findings show that post-solution confidence is consistently better calibrated and more discriminative than pre-solution confidence. Moreover, linear probes trained on hidden representations recover substantially richer confidence-related information than models explicitly verbalize. Building on this observation, we introduce future confidence distillation, which trains predictors operating on pre-solution hidden representations using teacher confidence estimates produced by post-solution correctness probes.
Despite requiring only pre-solution representations for inference, distilled predictors recover much of the calibration improvement achieved by post-solution confidence, remain highly sample efficient, and transfer across datasets within the same domain. Together, our findings demonstrate that confidence-related information evolves throughout the answering process and can be anticipated before answer generation is complete, enabling significantly more reliable yet low-cost confidence estimation.
Blogger's Review: This paper offers a novel perspective on confidence estimation by emphasizing its evolution, providing valuable insights for the practical application of large language models. Future confidence distillation enhances model calibration and demonstrates how to effectively leverage pre-solution information to optimize decision-making processes, which is significant for both practical utility and research implications.