Avatarl: Training language models from scratch with pure reinforcement learningtokenbender.com9 points·Gusarich··0 commentsOpen articleSaveView on HN