LLMs as Signal Detectors: Sensitivity, Bias, and the Temperature–Criterion Analogy

Jon-Paul Cacioli · 2026

Read LLMs as Signal Detectors: Sensitivity, Bias, and the Temperature–Criterion Analogy by Jon-Paul Cacioli (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-14893.txt.

Description

Cacioli casts language models as signal detectors and tests a temperature-criterion analogy. Calibration scores mix two skills: telling right from wrong, and leaning bold or shy. Signal detection theory splits those parts cleanly. Three models face 168,000 factual trials in a pre-registered design. Temperature should act like a payoff shift if the analogy holds. It breaks instead, moving sensitivity and criterion at once. Evidence spread stays unequal, with z-ROC slopes from 0.52 to 0.84. Instruct models skew more than the base model, Cacioli finds.

Catalog record

CREATORJon-Paul Cacioli
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39460
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.14893

Browse the full library · About the Tower of Babel