And GPT-3 can apparently already solve the Traveling Salesman problem which is wild to me. There's an increasing amount of reasoning that these new language models seem to produce - things like explaining a joke which has to be miles beyond Markov chain shenanigan territory.
I keep finding these developments fascinating and a bit scary. Can't help but think that we're in the honeymoon phase, as early signs are showing of things that are morally...grey. What happens when, I dunno, Facebook decides those billions of Whatsapp messages and voice memos might make an excellent training dataset?