as an outsider, what I find the most impressive is how long it took for people to realize this was a bubble.
Has been for a few years now.
Has been for a few years now.
The difference I see is that, conversely to websites like pets.com, AI gave the masses something tangible and transformative with the promise it could get even better. Along with these promises, CEOs also hinted at a transformative impact "comparable to Electricity or the internet itself".
Given the pace of innovation in the last few years I guess a lot of people became firm believers and once you have zealots it takes time for them to change their mind. And these people surely influence the public into thinking that we are not, in fact, in a bubble.
Additionally, the companies that went bust in early 2000s never had such lofty goals/promises to match their lofty market valuations and in lieu of that current high market valuations/investments are somewhat flying under the radar.
The promise is being offered, that's for sure. The product will never get there, LLMs by design will simply never be intelligent.
They seem to have been banking on the assumption that human intelligence truly is nothing more than predicting the next word based on what was just said/thought. That assumption sounds wrong on the face of it and they seem to be proving it wrong with LLMs.
However, even friends/colleagues that like me are in the AI field (I am more into the "ML" side of things) always mention that while it is true that predicting the next token is a poor approximation of intelligence, emergent behaviors can't be discounted. I don't know enough to have an opinion on that, but for sure it keeps people/companies buying GPUs.
That's a tricky metric to use as an indicator though. Companies, and more importantly their investors, are pouring mountains of cash in the industry based on the hope of what AI may be in the future rather than what it is today. There are multiple incentives that could drive the market for GPUs, only a portion of those have to do with today's LLM outputs.
If Burry could actually see a bubble/crash, he wouldn't be wrong about them 95%+ of the time... (He actually missed the covid crash as well, which is pretty shocking considering his reputation and claims!)
Ultimately, hindsight is 20/20 and understanding whether or not "the markers" will lead to a major economic event or not is impossible, just like timing the market and picking stocks. At scale, it's impossible.
What was the cost of the 16 missed predictions? Presumably he is up over all!
Also doesn't even tell us his false positive rate. If, just for example, there were 1 million opportunities for him to call a bubble, and he called 18 and then there were only 2, this makes him look much better at predicting bubbles.