Reproduce “The Trinity: Capability-Driven Recursive Self-Improvement for Multimodal 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
Weidong Tang, Ruiheng Wang, Jingxuan Zhang, Hexin Wang, Haoyuan Chen, Zhengfeiyang Zhang, Yixing Ma, Zichen Jiang and 13 more
The authors introduce Trinity, grouping tasks by shared capability requirements to build simulated practice worlds and refine Harnesses around a frozen model. At iteration 4, means rise from 39.06 to 73.16 across 28 Train settings and from 37.45 to 63.61 across 18 Unseen settings. Later results use prepared skills and Oracle selection, not demonstrating Router performance or model-weight improvement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.