Zichen Chen and Hugging Face open community voting for RSI Arena

Zichen Chen has partnered with Hugging Face to launch the next phase of RSI Arena. In the project, AI agents were provided with GPUs to independently select training data, write code, and conduct experiments. With round 1 of training completed, the platform now features head-to-head matchups where community votes serve as feedback for live continual learning.

Hugging Face Inference Endpoints power the arena by hosting each agent-trained model live, streaming user evaluation data directly back to the agents for their next research and training cycles.

The initiative is part of Open RSI, a project focused on open AI self-improvement. Chen stated that all models trained by the agents will be published on the Hugging Face Hub, and all project artifacts—including code, datasets, and complete agent research trajectories—will be open-sourced at the end of the experiment.