Rare Event Analysis of Large Language Models

Jake McAllister Dorman, Edward Gillman, Dominic C. Rose, Jamie F. Mair, Juan P. Garrahan · 2026

Read Rare Event Analysis of Large Language Models by Jake McAllister Dorman, Edward Gillman, Dominic C. Rose, Jamie F. Mair, Juan P. Garrahan (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-2602-06791.txt.

Description

Jake McAllister Dorman, Edward Gillman, Dominic C. Rose, Jamie F. Mair, and Juan P. Garrahan study rare events in large language models: behaviors that are far from typical yet highly consequential when they surface. Because model use is now enormous, events never seen during development are likely to appear prominently once deployed. The paper offers an end-to-end framework for finding, estimating, and characterizing them.

The implementation spans theory, efficient generation strategies, probability estimation, and error analysis, illustrated with concrete examples. It is designed to work across models and contexts, not just one system. The authors argue that systematic rare-event analysis should become a routine part of evaluating deployed models, alongside standard benchmarks that mostly measure average behavior.

Catalog record

CREATORJake McAllister Dorman, Edward Gillman, Dominic C. Rose, Jamie F. Mair, Juan P. Garrahan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE70 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-06791

Browse the full library · About the Tower of Babel