Comment by vivzkestrel

Stupid question but I had to ask, let us say instead of training anything, you start off with a completely different idea. You take a node, fluctuate its weights from -0.1 to 0.1, then you add another node with the same thing, then you add a 100 more nodes, 1000 more, another layer of 1000 more, and then you take the first math problem, tweak weights to get the right answer, take the second math problem, tweak weights to get the right answer, do this a billion times, maybe a trillion. will you eventually reach end up making a GPT?