analiza aiqbench
This arXiv paper proposes CoCHI, a method that locates and intervenes on specific attention heads inside LLMs to reduce overconfident wrong answers, rather than only rescaling outputs after the fact. The reported gains in ECE, NLL and Brier score across several models, with accuracy preserved on five multiple-choice benchmarks, are encouraging but limited to that setup. Readers should watch whether the approach generalises beyond multiple-choice tasks and whether the identified components hold up under broader testing.
Why Are LLMs Confidently Wrong? Correcting Overconfident Errors via Causal Head Intervention
arXiv:2610.34994v1 Announce Type: new Abstract: Modern large language models (LLMs) often exhibit overconfident errors, assigning high confidence to incorrect predictions and thereby undermining reliability in real-world use. Existing calibration methods typically rely on post-hoc adjustment or retraining, operating at the output level without addressing the underlying causes of miscalibration. We propose CoCHI, (Class-oriented Causal Head Intervention), a framework that improves calibration by identifying and intervening on internal components responsible for distinct confidence behaviours. By analysing predictions through correctness-confidence patterns, CoCHI enables targeted intervention on model internals, reducing overconfident errors while preserving correct predictions. Our approach provides a mechanistic perspective on calibration, linking reliability to identifiable structures within the model rather than treating it as an output-level issue. Experiments across multiple LLMs show consistent improvements in calibration metrics, including Expected Calibration Error, Negative Log-Likelihood, and Brier Score, while maintaining accuracy on five multiple-choice question answering benchmarks. These results suggest that effective calibration requires not only adjusting outputs, but also addressing the underlying causal mechanisms within the model.
Title:Why Are LLMs Confidently Wrong? Correcting Overconfident Errors via Causal Head Intervention
Abstract:Modern large language models (LLMs) often exhibit overconfident errors, assigning high confidence to incorrect predictions and thereby undermining reliability in real-world use. Existing calibration methods typically rely on post-hoc adjustment or retraining, operating at the output level without addressing the underlying causes of miscalibration. We propose CoCHI, (Class-oriented Causal Head Intervention), a framework that improves calibration by identifying and intervening on internal components responsible for distinct confidence behaviours. By analysing predictions through correctness-confidence patterns, CoCHI enables targeted intervention on model internals, reducing overconfident errors while preserving correct predictions. Our approach provides a mechanistic perspective on calibration, linking reliability to identifiable structures within the model rather than treating it as an output-level issue. Experiments across multiple LLMs show consistent improvements in calibration metrics, including Expected Calibration Error, Negative Log-Likelihood, and Brier Score, while maintaining accuracy on five multiple-choice question answering benchmarks. These results suggest that effective calibration requires not only adjusting outputs, but also addressing the underlying causal mechanisms within the model.
Artificial Intelligence (cs.AI)
arXiv:2610.34994 [cs.AI]
(or arXiv:2610.34994v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.34994 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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 · 267 cuvinte
Subiecte conexe
Calibrating Overconfidence Without Sacrificing Confidence: Probe-Conditioned Head Intervention for LLMs
arXiv · 10 iun. 2026
What Shared Prefixes Hide: Trajectory Dropout for On-Policy Distillation
arXiv · 29 sept. 2026
Wired for Overconfidence: A Mechanistic Perspective on Inflated Verbalized Confidence in LLMs
arXiv · 21 iul. 2026
Know More, Know Clearer: A Meta-Cognitive Framework for Knowledge Augmentation in Large Language Models
arXiv · 9 iun. 2026