Reproduce “World Intelligence Agent: Unifying World Modeling and Action Learning for Digital Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Weicheng Meng, Shuo Lu, Guangzeng Han, Yuanfa Li, Yijun Liu, Kuangpu Guo, Jingyang Qiao, Yu Cheng and 11 more
WIA presents a world model that predicts a digital agent’s next response from its interaction history and action. It uses a shared action–observation interface and evaluates predictions across seven domains, with out-of-domain tests limited to sources or benchmarks within those categories. On WIA-Bench, its 122B-A10B model scores 57.35 versus 53.19 for Kimi-K3 under the judge and a different training recipe.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.