This is the most grounded take on model scaling I’ve seen in a while. Small models doing real work makes way more sense for agent-based systems.

→ View original post on X — @godofprompt