One would expect people to just use FPGAs (silicon that can be reprogrammed at the gate level) instead of ASICs (silicon specifically burned for the application) and just get 1/10th the speed and the ability to reprogram for frontier models, since 1,700 tokens per second is still well above what is needed for most applications.
One would expect the moment these guys' architecture is reverse-engineered the FPGA-wielders will reduce their stock price to zero, but then again, there's a lot of cash to burn in AI startup land.
It seems like datacenters are the front that could cripple AI in this country. I keep hearing about votes against building new centers in rural areas. A friend mentioned that a proposed data center would need 1.5 the volume of water for what is currently used by the small city’s residents. The idea of losing or polluting water has the potential to get the MAHA moms out in force against it.
As laid out in yesterday's article, if the powers that be decide that they want powerful AI they will construct powerful AI regardless of the opinions of the plebs.
One would expect people to just use FPGAs (silicon that can be reprogrammed at the gate level) instead of ASICs (silicon specifically burned for the application) and just get 1/10th the speed and the ability to reprogram for frontier models, since 1,700 tokens per second is still well above what is needed for most applications.
One would expect the moment these guys' architecture is reverse-engineered the FPGA-wielders will reduce their stock price to zero, but then again, there's a lot of cash to burn in AI startup land.
From a business point of view, the locked model is an advantage.
Nobody wants to pay for software, but with this as long as models keep iterating, you can sell hardware.
Physical product.
As to whether it'll actually play out that way?
Beyond this teller's capabilities.
Recreating the old churn of rapidly obsolete tech hardware would definitely get some suits salivating though.
...
It's still crazy how effectively small a history computers have.
Very few foresaw where we are now.
Few, but not none.
It seems like datacenters are the front that could cripple AI in this country. I keep hearing about votes against building new centers in rural areas. A friend mentioned that a proposed data center would need 1.5 the volume of water for what is currently used by the small city’s residents. The idea of losing or polluting water has the potential to get the MAHA moms out in force against it.
Maybe.
As laid out in yesterday's article, if the powers that be decide that they want powerful AI they will construct powerful AI regardless of the opinions of the plebs.