Comment by dougmwne

It seems quite clear that we are just a few years from being able to wire up less than 20k of nvidia hardware and produce a GPT-4 class model. All the methods and datasets are out there. Plenty of pre-trained base model weights too. The horse is out of the barn.