It can reason and it can extract and break down concepts in complicated material. It does a really good job imitating voice and style if provided with a sample. It doesn’t often make the kinds of dumb mistakes GPT-3.5 did, and I have yet to see it hallucinate. It also doesn’t require such elaborate prompt engineering (although it helps).

I’ve used it for work daily over the last few weeks, and more than once I’ve had to get up and pace around the house for a bit to process that this thing is probably smarter than me (although I still edge it out in some really in-the-weeds stuff in my particular area of expertise, but it won’t take long for it to catch up).

Also, even if it’s not quite as smart or talented as humans in whatever fields, consider that neither are self-checkouts, but they are ubiquitous because they do a minimally satisfactory job while being a lot cheaper. That’s all it takes.