A dollar auction works as follows: it’s a continuous auction where the highest bidder gets the dollar while everyone pays their last bid.
At the start this seems like a great deal: you bid 5 cents hoping to walk away with a nice 20x payday. Then another player bids 10 cents; still an awesome deal so you counter. As the bids are approach $1 you would think the bidding would settle down but that’s not the case.
The key rule to remember is that everyone pays their bid regardless of whether they win the dollar. So consider what happens when your 90 cent bid gets trumped by a 95 cent bid. Surely it would be dumb of you to bid a $1 to win a $1. But at least then you’d break even instead of losing 90 cents. And surely it would be even dumber to bid $1.05 but the same logic holds for your newly trumped opponent: at least they would only lose 5 cents rather than $1. And so it goes until you either get wise and cut your losses or have them cut for you by running out of capital.
So about generative AI… for the past few years I’ve been a LLM skeptic. What I couldn’t understand is the vast sums of money being invested without a clear path to a return. With the facade of AGI dropping this becomes even less understandable. To abuse several metaphors: if you keep in the game long enough to build a city of geniuses in a data center then the dollar that’s being auctioned off suddenly transforms into a trillion dollar coin1 and everything makes sense. But aside from that: where’s the ROI?
The zeitgeist seems to be equating the LLM era with the railroad era. Lots of money flowed into the railroads, they spectacularly went bust, but railroads ended up being useful. From the outside the dynamic seems reasonable but what I am curious about is the motivations of the railroad investors. Curiously it seems like retail investors were part of the frenzy. Given the pile into NVidia history seems destined to rhyme.
-
presumably with Elon’s face on it ↩︎