Sung Choi compares prompting AI models to managing a junior analyst: you need to give detailed, specific instructions or the work falls apart. The skill translates directly from learning to research on Google in the early 2000s. Better prompts come from iteration and studying how others use these tools.
Chapters
I remember in college, back in the early 2000s, there was a class where they taught you like how to research better on Google. And today the equivalent of it is, uh, like, you know, how to ask better questions to Claude or ChatGPT, right? So I think of AI as an analyst and as a junior analyst that's really, really good at researching, um, and is extremely thorough, but very junior. So I have to describe, uh, give it very great descriptions as to exactly what I'm looking for. Otherwise I'm going to get subpar output. That's helped me kind of understand how I want to ask questions and how I want to do it. And then, you know, between, you know, online research and like newsletters I'll subscribe to about AI and things like that, like I keep getting like little tidbits of like better ways I can contextualize what AI is giving me and then how to get it to answer. Over time and over a lot of iterations, I started understanding, wow, here's all the amazing things that it can do for me, but I just need to make sure that I answer, I ask things in a very specific way.