People often correct bad generation and I think such feedback is valuable. eg "Training Language Models with Language Feedback at Scale" https://arxiv.org/abs/2

People often correct bad generation and I think such feedback is valuable. eg "Training Language Models with Language Feedback at Scale" https://arxiv.org/abs/2303.16755 https://arxiv.org/abs/2303.16755