analiza aiqbench
This research addresses a critical gap in LLM reliability by shifting focus from single-response confidence to query-level capability calibration, offering a more accurate metric for real-world deployment. The study demonstrates that current response-level methods fail to capture the stochastic nature of modern decoding processes, leading to misaligned confidence estimates. By introducing a new framework that aligns confidence with expected accuracy, this work provides essential tools for optimizing inference budgets and predicting success rates in complex tasks.
On Calibration of Large Language Models: From Response To Capability
arXiv:2602.13540v2 Announce Type: replace Abstract: Accurate confidence estimation is critical for reliable use of large language models (LLMs). Prior work on LLM calibration largely focuses on response-level confidence, which estimates the correctness of a single generated output. However, this formulation is misaligned with many practical settings where the central question is how likely a model is to solve a query overall. We show that this mismatch results from the stochastic nature of modern LLM decoding, under which single-response correctness fails to reflect underlying model capability. To address this issue, we introduce capability calibration, a new evaluation framework for measuring how well query-level confidence aligns with a model's expected accuracy on individual queries. We formally distinguish capability calibration (CC) from response calibration (RC) and show that the two differ both theoretically and empirically. We further show that CC is better suited than RC to applications like pass@k prediction and inference budget allocation. Finally, we evaluate common confidence estimation methods to understand the practical feasibility of CC.
Abstract:Accurate confidence estimation is critical for reliable use of large language models (LLMs). Prior work on LLM calibration largely focuses on response-level confidence, which estimates the correctness of a single generated output. However, this formulation is misaligned with many practical settings where the central question is how likely a model is to solve a query overall. We show that this mismatch results from the stochastic nature of modern LLM decoding, under which single-response correctness fails to reflect underlying model capability. To address this issue, we introduce capability calibration, a new evaluation framework for measuring how well query-level confidence aligns with a model's expected accuracy on individual queries. We formally distinguish capability calibration (CC) from response calibration (RC) and show that the two differ both theoretically and empirically. We further show that CC is better suited than RC to applications like pass@k prediction and inference budget allocation. Finally, we evaluate common confidence estimation methods to understand the practical feasibility of CC.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2602.13540 [cs.CL]
(or arXiv:2602.13540v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2602.13540 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 259 cuvinte
Subiecte conexe
[AINews] High Taste Testers Ep. 1
Latent Space · 17 oct. 2026
[AINews] The Future of Latent Space
Latent Space · 25 sept. 2026
[AINews] Opus 5.5 is good at explainer videos
Latent Space · 29 sept. 2026
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · 13 oct. 2026