Reproduce “OfficeTown: Generative Organizations for Evolving Agent Intelligence” 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
Hongchen Wei, Daohong Ying, Zhi Hong, Aidan Wu, Bei Liu, Ruichun Ma, Yifan Yang, Qi Dai and 5 more
OfficeTown generates professional-work tasks by simulating organizations, where decisions, files, feedback, and open questions persist under role-based permissions. It extracts tasks from requests and assignment-time materials and uses these environments for evaluation and training. On 79 GDPval tasks, Qwen3.5-9B’s rubric score rose from 65.75 to 73.65 after SFT and OPD; live-web-search tasks were excluded, and the result does not isolate the cause.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.