Reproduce “Measuring AI Agents as Teammates with Collaborative 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
Sean Wu, Aneesh Pappu, Jiaru Zou, Yubin Kim, Yuncong Mao, Hongjian Zhou, Junchi Yu, Jieneng Chen and 1 more
The benchmark measures teammate behavior by comparing tasks solved alone with scripted teammate-problem tests. Its score averages eight dimensions, using only tasks passed alone. Gemini 4 Argon scores 82.2/100 at high reasoning effort. In pressured conflicts, frontier models find the best deal in 14% of runs; concessions average 95% of best joint points. Prediction on other team tasks remains untested.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.