How long will the next AI winter last? Any predictions?

I think this one may even be longer than the last one because this cycle was entirely due to the existence of GPUs and NVIDIA. There hasn't been a single fundamental innovation. Neural networks are ancient and are unlikely to be the holy grail of intelligence after multiple decades of failed attempts. It's like running an old game on an overclocked modern desktop and claiming you invented a new game. Since we are at the absolute peak of hype now (maybe a little past the peak already), this should be the start of the bear market for AGI. The real question is if Lisp will make a comeback in the aftermath.

>>1
I strongly disagree. Long short-term memory works amazingly well to solve increasingly complex problems we used to program for. Take a look at AIs playing games, GPT-2, Transformer writing anything from genius anti-semitism to homosexual jokes, and the incipient but game-changing business created by AIs able to create music.

Let me take the AI-composed music as an example. Here: https://www.youtube.com/watch?v=TqH6ionF94I after learning Chopin, and here: https://www.youtube.com/watch?v=2kuY3BrmTfQ after learning Vivaldi. Sure, it sucks. It can't compare to Chopin's genius, who was one of the very best composers ever, highly regarded and IMO still grossly underrated. But the key is not that it's amazing. It's that it's good enough. Good enough to be used by a low or mid-budget Netflix show. Good enough for most videogames. Far too good to be used by today's shitty TV programmes, let alone background music in company videos, facilities, speeches, and so on. It's good enough and it works.

I'm talking about businesses like this: https://www.aiva.ai/

If this is not revolutionary, tell me what is. I don't think winter is coming.