Reproduce “Model What Matters: Decision-Guided Selective World Modeling for Language 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
Maolin Ran, Minda Chen, Di Huang, Wei Xiong, Fei Huang, Weiwen Liu
DGS-WM trains language agents with auxiliary world-model supervision that selects observation units supporting the next response and weights selected targets by model confidence. On Qwen3-1.7B, it reaches 22.06% multi-hop QA exact match, versus 17.71% for GRPO+ECHO and 18.88% for GRPO alone. This result is a benchmark comparison; response support is correlational, not causal.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.