Prompt Engineering可以收皮

啱啱睇完個langchain嘅tutorial，坊間主流媒體講乜鬼做prompt engineer可以有幾多十萬美金人工真係得啖笑。langchain先係LLM嘅發展方向，學乜鬼prompt engineering嘅可以收皮。

prompt engineering收乜野皮 定你想講prompt engineer收皮
langchain啲few-show in-context learning prompt template唔算prompt engineering?
將knowledge base index完咗之後query塞落prompt又唔算prompt engi嘅retrieval augmented generation?
啲chain又唔算chain-of-thought?
行agent又唔算行緊ReAct logic?

仲有都唔知你係咪真係用過langchain寫過野
成舊野根本就好撚bloat 啲syntax又verbose
根本就over-engineered 真正production根本唔會想用
啲想做到嘅llm flow明咗prompt engi concept之後自己用code砌好過用langchain

基本上你同LLM互動就有prompt engineering成份㗎喇,收乜嘢皮

即係你以前學programming時，一定有學過點寫quicksort，乜鬼red-black tree，讀完書後你有冇真係落手落腳寫過丫？乜鬼都係call library call API。寫個prompt去query個LLM，遲早都係變成走去call唔同嘅API。

如果你prompt engineering冇做好嘅話你行langchain佢都係會畀啲屎你

咁你會唔會claim quicksort同red-black tree可以收皮先？
嗰樣野比人abstract咗detail implementation可以唔洗識原理都識用同嗰樣野可以收皮係兩回事