Flutterby™! : AI spending

Saved off here because I've been writing this same sentiment in a few places recently. In this latest case, a friend who commented that "AI" spending was gonna be $400B this year.

So depending on how you allocate money to what, LLMs have sucked up $1.5-2T since 2018. Current revenues are, very roughly, about $60B/year, at huge operating losses. If we figure that investment should have, I dunno, call it 5% return? So even without making up for lost time, after 7 years ya kinda think there'd need to be $75-100B in profit, not just revenue.

And, sure, it's a moon shot, Kennedy gave his "We choose to go to the Moon" speech September of '62, Apollo 11 landed July of '69 (oh, look, 7 years), but the whole LLM approach feels very much like it's running out of steam.

And AI hardware depreciates like no other purchase, and it's gonna be hard to re-purpose data centers for something useful.

Sure, there's a lot of really cool stuff coming out of neural net research, some of the stuff that's happening in image recognition is extremely cool, but... The crash is gonna suuuuuuuck. Like the US in the 1890s in the levels of suck.