The Allen Institute for AI (Ai2) recently released what it calls its most powerful family of models yet, Olmo 3. But the company kept iterating on the models, expanding its reinforcement learning (RL) ...