推荐not_a_bot_4sho · 2026-06
“if you're going to only read one paper here, maybe read this one but know there has been a lot of follow up with more modern models.”
只读一篇的话可以读这篇,但后来有更多针对新模型的跟进研究
自动生成专家身份描述来提升 LLM 回答质量的提示方法
论文提出 ExpertPrompting:先用上下文学习为每条指令自动合成详细、定制的专家身份描述,再让 LLM 以该背景作答。作者用这种增强提示方式,借助 GPT-3.5 生成新的指令数据,训练出开源对话助手 ExpertLLaMA。基于 GPT-4 的评估显示专家数据质量显著高于普通回答,ExpertLLaMA 优于现有开源对手,达到原版 ChatGPT 能力的 96%。
HN 评论者把它当作专家角色提示的起点,建议在可行时使用,但也提醒后续已有更多针对新模型的研究,结论可能过时。
“if you're going to only read one paper here, maybe read this one but know there has been a lot of follow up with more modern models.”
“there's actually quite a bit of research in this field, here's a couple:”
“See examples in ExpertPrompting; They certainly do say things like”
“use expert prompting[0] when possible.”