Language models contain built-in welfare axis that training activates

Original: How's it going? Reinforcement learning in language models recruits a functional welfare axis

Source: arxiv.org ↗