TimeCausality: Evaluating the Causal Ability in Time Dimension for Vision Language Models

Zeqing Wang Shiyuan Zhang Chengpei Tang Keze Wang · 2025

Read TimeCausality: Evaluating the Causal Ability in Time Dimension for Vision Language Models by Zeqing Wang Shiyuan Zhang Chengpei Tang Keze Wang (2025) 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-2505-15435.txt.

Description

Zeqing Wang, Shiyuan Zhang, Chengpei Tang, and Keze Wang build TimeCausality, a benchmark for causal reasoning over time in vision-language models, in work from twenty twenty-five. People read irreversible change, like fruit rotting or ageing, with ease. Such reading needs world knowledge, not just event order spotting. Whether vision models share that skill was largely untested. The new benchmark isolates temporal causality as its target skill.

Top open models match closed models on standard visual question tasks. On TimeCausality those open models lag closed rivals by a wide gap. Even the strongest closed model drops sharply on this benchmark. The gap flags temporal causality as a weak spot across the field. Future model training should fold in change-over-time reasoning. Open developers gain a clear target to close next.

Catalog record

CREATORZeqing Wang Shiyuan Zhang Chengpei Tang Keze Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-15435

Browse the full library · About the Tower of Babel