A popular opinion that we generally see after AI is any job is doable now, inlcuding and especially software. While that might hold some truth to it, but not entirely. Thats exactly what this article argues about written by Sean Goedecke, a staff engineer at GitHub who works on Copilot and other AI products. Give a mathematician and a doctor the same model and the same math problem, and the mathematician gets far better results. The model didn't change, the person asking did. I agree completely, domain knowledge is what turns an LLM from a generalist into a real collaborator.
LLMs reward expertise
Everyone talks to the same models, but domain knowledge decides what you get out of them: experts can steer, push back, and spot what looks wrong.
- llms
- expertise