Foundations of Global Consistency Checking with Noisy LLM Oracles

Paul He, Elke Kirschbaum, Shiva Kasiviswanathan · 2026

Read Foundations of Global Consistency Checking with Noisy LLM Oracles by Paul He, Elke Kirschbaum, Shiva Kasiviswanathan (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-2601-13600.txt.

Description

Paul He, Elke Kirschbaum, and Shiva Kasiviswanathan study how to check if a large set of facts fits together in 2026. Tasks like fact checks, short summaries, and knowledge bases all need facts that do not clash. Large language models can judge small groups of facts for clashes. But their calls are noisy, and pair-by-pair checks can miss clashes that span the whole set.

The team lays the base math for global checks with such flawed judges. They ask what can still be known when each small check may err. The work maps the limits of trust in model-based checks. It also points to where extra care or proof is a must.

Catalog record

CREATORPaul He, Elke Kirschbaum, Shiva Kasiviswanathan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-13600

Browse the full library · About the Tower of Babel