There were so many AI experts trying to beat Go that they saw many, many problems. So the lesson is that computers can get really good at one thing, providing that there are clear rules.

I think that Generative AI will crash and burn soon. I mean, look at ChatGPT. You need top GPUs to work a long time using a huge network that is not even trained on all the text on the Web. You could maybe increase the size of the network a thousand times, but you will need more than a thousand times more GPUs. Much more. And at inference time you still need the parameters. I am afraid it will not be enough to accommodate multimodal abilities and larger context windows.