Comment by datacynic

Writing documentation for LLMs is strangely pleasing because you have very linear returns for every bit of effort you spend on improving its quality and the feedback loop is very tight. When writing for humans, especially internal documentation, I’ve found that these returns are quickly diminishing or even negative as it’s difficult to know if people even read it or if they didn’t understand it or if it was incomplete.