RL Post-training Amplifies Pretraining Behaviors in Language Models

Best AI papers explained - RL Post-training Amplifies Pretraining Behaviors in Language Models