Differentiable Conformal Training for LLM Reasoning Factuality

Nathan Hittesdorf, Marco Salzetta, Lu Cheng · 2026

Read Differentiable Conformal Training for LLM Reasoning Factuality by Nathan Hittesdorf, Marco Salzetta, Lu Cheng (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-2604-20098.txt.

Description

Hittesdorf, Salzetta, and Cheng tackle false claims in multi-step language model thought with a smooth take on conformal fact checks. Old coherent fact-check tools need hand-made scorers that, at high trust levels, throw out nearly 60 percent of true claims. Their smooth Coherent Factuality tool learns better scorers while keeping the first tool's pledges intact. On two thought benchmarks it lifts kept claims by up to 141 percent with trust intact, a step toward conformal tools users can trust.

Catalog record

CREATORNathan Hittesdorf, Marco Salzetta, Lu Cheng
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE78479
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.20098

Browse the full library · About the Tower of Babel