I spent some time experimenting with AI art. He's right that it doesn't look right somehow most of the time. But in my case, if I started with a photograph I took and extracted its silhouette lines as the basis for the image I was generating, I got something a lot harder to distinguish from manually generated art.

He's right about the motives of tech Bros and executives here. But I don't think he's right about where AI will eventually go because I think it will become pervasive because it almost already is.

It's ridiculously optimistic to think you can run a 1 trillion parameter model on a phone anytime soon. So those data centers have a power generation problem and their limitation will be the extent to which the power generation problem is solved.

I just don't think AI is going to crash. But if you look at ridiculous valuations and PE ratios higher than 100 or so, timber! But crashing to 20 to 50 (CSCO after the crash from 200) is just the market pulling its own head out of its butt so I suppose there's the bubble. Good luck, however, figuring out when. Also, CSCO is still here. He really ought to look around at how many people have outsourced their thinking and writing to large language models already.

I'm a bit reminded of this Outer Limits episode minus the BCI:

https://www.imdb.com/title/tt0667957/plotsummary/

I don't think senior talent will get replaced anytime soon, but I also think that one or two fundamental innovations down the road on par with the Attention is All You Need paper will make AI dramatically better at tasks that don't require embodiment. Those innovations seem to occur every couple years across the sector. But again, good luck. predicting when and how.

See also: https://medicushcs.com/resources/the-radiologist-shortage-ad...

Before we fire all the radiologists, we'll have to centaur the crap out of them just to meet demand. But that doesn't fit his narrative so he left that out because I'm sure he's aware of it.