supervision signal

hard label (1) large language model (1) performance gap recovered (1) ai alignment (1) weak-to-strong generalization (1) soft label (1) overfitting mitigation (1) generative model (1) model alignment (1) few-shot learning (1)