Comment by layer8

The parent comment didn’t say anything about expecting the LLM output “to be line-by-line what you think is most appropriate as a human”?