AI zero to one was moonshot hard. AI 1 to n is trivial and only limited by your compute budget. I’ve never seen so many people acquire basic mastery of a tech so fast. Very smart people yes but not Einsteins. I’d guess top decile of devs can master this in a few years. And GPT itself will help do most of it.

Blogger at ribbonfarm.com, running summerofprotocols.com, host of /gloom ☀️

doing deep learning for a master's rn, using PyTorch. In some ways it feels like learning to program all over again, in other ways it's just regular programming

I think I'll be good in a few years (or less). And then, they say LLaMa when fine-tuned is like GPT-3.5, and it's open source...