6 ms·Avatarl: Training language models from scratch with pure reinforcement learning9 points by Gusarich 1y ago