Reading Calibrated Uncertainty from Language Model Trajectories

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Eusebi, Aliai, Herzog, Alexander, Liang, Xiaoyu, Vasek, Marie, Mariconti, Enrico, Cavallaro, Lorenzo
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913153943601152
author Eusebi, Aliai
Herzog, Alexander
Liang, Xiaoyu
Vasek, Marie
Mariconti, Enrico
Cavallaro, Lorenzo
author_facet Eusebi, Aliai
Herzog, Alexander
Liang, Xiaoyu
Vasek, Marie
Mariconti, Enrico
Cavallaro, Lorenzo
contents The maximum softmax probability (MSP) represents a default approach when evaluating uncertainty quantification for language model generation with structured output. Although cheap, it is often miscalibrated. Methods that probe the model's internal activations feed raw hidden states into opaque classifiers, reading activations as static snapshots and leaving implicit the layer-wise trajectory by which a representation is formed. Yet, similar endpoints can arise from very different paths, and how evidence accumulates, reinforces, or reverses across depth might reveal uncertainty that final probabilities obscure. We extract eleven scale-invariant geometric features, tracing the cumulative path of per-layer MLP updates, and feed them to a sparse linear probe. The probe outperforms MSP under selective abstention, with gains scaling with baseline miscalibration up to 21 AURC points. Because every feature has a closed-form geometric meaning, the probe's coefficients trace how and where along depth errors take shape -- which layers commit prematurely, which contradict the running state, where trajectories drift away from their endpoint.
format Preprint
id arxiv_https___arxiv_org_abs_2605_22864
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Reading Calibrated Uncertainty from Language Model Trajectories
Eusebi, Aliai
Herzog, Alexander
Liang, Xiaoyu
Vasek, Marie
Mariconti, Enrico
Cavallaro, Lorenzo
Machine Learning
The maximum softmax probability (MSP) represents a default approach when evaluating uncertainty quantification for language model generation with structured output. Although cheap, it is often miscalibrated. Methods that probe the model's internal activations feed raw hidden states into opaque classifiers, reading activations as static snapshots and leaving implicit the layer-wise trajectory by which a representation is formed. Yet, similar endpoints can arise from very different paths, and how evidence accumulates, reinforces, or reverses across depth might reveal uncertainty that final probabilities obscure. We extract eleven scale-invariant geometric features, tracing the cumulative path of per-layer MLP updates, and feed them to a sparse linear probe. The probe outperforms MSP under selective abstention, with gains scaling with baseline miscalibration up to 21 AURC points. Because every feature has a closed-form geometric meaning, the probe's coefficients trace how and where along depth errors take shape -- which layers commit prematurely, which contradict the running state, where trajectories drift away from their endpoint.
title Reading Calibrated Uncertainty from Language Model Trajectories
topic Machine Learning
url https://arxiv.org/abs/2605.22864