It's even easier now than 12 years ago when I started. Better tooling, more efficient pre-trained models, and the whole prompting paradigm that makes a month l

It's even easier now than 12 years ago when I started. Better tooling, more efficient pre-trained models, and the whole prompting paradigm that makes a month long task doable in a day. You don't have to label as much as before, or not at all. When you train small models on small datasets you get very bad out-of-distribution results, but when you use these LLMs they have already seen everything on the web so they are not as often OOD.