Two Novel AI Systems for Multi-Agent Cooperation and GPU Kernel Optimization

According to the first arXiv preprint, the researchers have introduced a new framework called “AsymPuzl” for studying multi-agent cooperation in controlled settings, rather than open-ended role-play scenarios. AsymPuzl is described as a “minimal but expressive two-agent puzzle” that could help evaluate how AI agents coordinate and solve problems together.

The second arXiv paper presents “Astra