Key Info

RSI Arena is running a new round in partnership with Hugging Face where AI agents autonomously choose data, write code, and run experiments to train better models, with community votes fed back into the loop.

Highlights

  • Round 1 of agent-driven training is complete; the community now acts as judge in a head-to-head model comparison
  • Agent feedback loop is streamed live, enabling continual learning from user votes
  • Hugging Face Inference Endpoints serve every agent-trained model live during the arena
  • Agents receive community feedback and launch new experiments and training runs in response