ParentFull threadbenlivengood·The comparison between base pretrained models and the RLHF production/helpful models suggests that it's likely reinforcement learning which is developing the introspection.View on HN