Reproduce “Language Models as AI Research World Models” 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

Zijun Wang, Zewen Liu, Minhua Lin, Zhaotian Weng, Zhan Shi, Bing He, Yisi Sang, Dakuo Wang and 5 more

The authors test a language model that predicts intervention outcomes from a target research setting and records of earlier experiments supplied in-context, without fine-tuning. Across five autoregressive pretraining targets, cross-environment records raise mean Spearman correlation from 0.63 to 0.73 for unseen interventions when the target contributes no records. Fixed-budget selection improves for four of five targets, not Marin.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.