Qinyu LuoBYTEDANCE SEED
Let’s talk
All research

ACL 2026 Conference paper

Current Agents Fail to Leverage World Model as Tool for Foresight

An empirical study of whether vision-language agents can use external world models to anticipate action outcomes.

Cheng Qian, Emre Can Acikgoz, Bingxuan Li, Xiusi Chen, Yuji Zhang, Bingxiang He, Qinyu Luo, Gokhan Tur, Dilek Hakkani-Tür, Yunzhu Li, Heng Ji

Fig. 1 — Original world model as a tool framework.Original source ↗
World Model as Tool
World model as a tool framework, with an agent choosing real actions or simulated feedback.

Fig. 1 — Original world model as a tool framework.

01 / The question

Does access to a simulator help an agent make better decisions about the future?

02 / The approach

World Model as Tool

The study evaluates world-model use across agentic and visual question-answering tasks, then analyzes when agents simulate and how they interpret the results. Its findings motivate better policies for invoking simulation and integrating predicted outcomes into decisions.

Read the full paper

03 / Citation

Cite this work

Select the citation to copy it manually.

04 / Versions & records

Publication record

Earlier or alternate titles

  • World Model as Tool: An Empirical Study on Agent Foresight Governance
2 Google Scholar records

READ HERE · 38 PAGES

Full paper

Document

Loading document…

Scroll through every page. Select text to copy, or use search to find a passage.

More research

View all
ACL 2025

EscapeBench

Room-escape environments for studying creative problem-solving, implicit goals, and unexpected uses of familiar tools.

arXiv 2026

One-Shot On-Policy Distillation

Investigating why on-policy distillation can recover much of full-data training’s gain from a single query.

GET IN TOUCH

Let’s talk.

For research, collaboration, or a conversation. Email me or connect on LinkedIn and X.

Open email app

You can also select and copy the address above.

LinkedInXGitHubGoogle Scholar