A new study from Cornell researchers just found that increasing the number of AI agents collaborating on a problem can significantly improve performance. Notably: A 13B parameter Llama model beat a 70B version on some tasks by scaling the number of agents up.

→ View original post on X — @rowancheung,