It takes years to decades for new technologies to achieve their full effect. Factories based on steam had to be rebuilt around eletricity. That took decades. It will take time to rebuild large operations around AI.
In fact I consider this rather obvious and borderline self-evident to a careful thinker.
However, the AI stocks are priced as if they are going to reap all the benefits, if not today, then perhaps even about a month or two ago. They are priced as if the current incumbents are the inevitable inheritors of all the promises, because the AI world is going to move so quickly to reap all this supposed value that there isn't even time for other companies to move in and displace nVidia. They are priced as if that productivity curve has already started to shoot up and it's just the beginning.
If the curve isn't even budging, then the AI world is going to need to be seriously, seriously repriced. If AI is going to change the world, but in five years rather than five months, and gradually rather than all at once, then the time value of money implies that the value to be captured in today-dollars is much, much smaller. In five years, nVidia's AI may be getting stomped by a startup that hasn't even been founded yet; I'd give it low odds, but meaningfully non-zero odds.
Whether I'm an "AI skeptic", I don't know. But I absolutely am an AI bubble skeptic. I haven't seen any evidence that the value creation is emerging at anything like the rate the bubble pricing implies. It'll probably arrive, but a lot more gradually than the market is currently believing.
Yeah, there's actually a huge amount of metaphorical territory--if sparsely populated--between "optimistic about AI in a general sense" versus "believing all of today's market/startup hype".
However it feels like the window has been dragged so far to the hype-side that anything less than breathless stock-buying while waving a The Singularity Is Near sign will get you a "pessimist" or even "Luddite" label.
When you have an engine that consistently provides torque better and cheaper than humans or animals, you can literally build a factory around it. ML tools are currently something you occasionally try in case it gets you on track faster, but you need to have a fallback to rely on. There is nothing consistent about it.
At least yet.
Apparently you haven't worked with many humans...
this goes the other way too, AI that trains on data from humans, well you're going to get the same sort of reliability haha
We get the best results from neural networks on closed systems with an infinite supply of training data (think how successful alpha zero was at chess, it simply trained by playing games if I'm not mistaken.)
Neural networks are an incredible tool. AI as a chatbot trained on human input is going to probably just simulate a lower quality human being that hallucinates more than normal lol
I work in AI and I think he’s right to a degree.