The fibre in the ground from the dotcom bubble was still used… eventually.
Most of the 100’s of billions of dollars of GPUs that have just been sitting on shelves in warehouses and will continue to sit there for years even if there isn’t a crash will be nearly worthless (relative to the prices paid) whether they try to mass liquidate at a small fraction of the purchase price to pay back some of their 10%+ interest loans they used to buy them or they sit around for a decade until there’s a place to put them and at that point they could be so inefficient, they’re not worth the money to install them and power to run them.
Yeah, chips are more of a perishable good than any kind of durable infrastructure, and at the end of the day, the models themselves have no durable value because information is never truly scarce, regardless of how they try to make us pretend it is or poison the well with lies. There’s no magic solution to making LLMs actual AGI or even effective tools outside of their narrow scope. Not only that, the way they’re using it is extremely reckless, deeply inefficient, and supremely cost intensive, on a level that makes the Manhattan Project look wholesome. There isn’t going to be a “last man standing” to win, it’s just going to go up like flash powder and take everyone in the dogpile with it.
The fibre in the ground from the dotcom bubble was still used… eventually.
Most of the 100’s of billions of dollars of GPUs that have just been sitting on shelves in warehouses and will continue to sit there for years even if there isn’t a crash will be nearly worthless (relative to the prices paid) whether they try to mass liquidate at a small fraction of the purchase price to pay back some of their 10%+ interest loans they used to buy them or they sit around for a decade until there’s a place to put them and at that point they could be so inefficient, they’re not worth the money to install them and power to run them.
Yeah, chips are more of a perishable good than any kind of durable infrastructure, and at the end of the day, the models themselves have no durable value because information is never truly scarce, regardless of how they try to make us pretend it is or poison the well with lies. There’s no magic solution to making LLMs actual AGI or even effective tools outside of their narrow scope. Not only that, the way they’re using it is extremely reckless, deeply inefficient, and supremely cost intensive, on a level that makes the Manhattan Project look wholesome. There isn’t going to be a “last man standing” to win, it’s just going to go up like flash powder and take everyone in the dogpile with it.