They do degrade physically, but the bigger thing is they stop being competitive quickly. Each year or so we see doubling of GPU speeds for the same amount of power.
If you build a 100MW data center with GPU compute and three years laster a new data center opens with the same cost for GPUs and same electricity cost you do, but can do twice as much compute, you quickly lose business unless the market is just so constrained customers can't afford to be picky. But the moment there's slack in the market you'll see major migrations off of providers that have the same cost but half, or quarter of the same performance.
So when you see someone talking about GPUs fully deprecating in value in 1-3 years this is what they're talking about. Right now it's not a big deal because there's no slack in the market. But once there is, the bottom will drop out.
ANY online device? Even assuming AI can find vulnerabilities in every operating system, there's no indication that this is actually true beyond a "here's how it could work"
This is the same nonsense that lead to article saying researchers had created a wormhole when all they had done was draw one.
I have a microcontroller with an ROM disk (i.e., physically read only). You're telling me that an AI can find a way around the physics of not being able to mutate ROM and exploit it?
I mean, if it's online it has a network/wireless card and a TCP stack along with at least some amount of RAM, so yea, in theory unless the programming is perfect it could be exploited. Now, it's not going to be used to run AI, but could very well get used in a DDOS or something like that.
I'm relieved my entire network consists of old computers and small, low power network devices. The only dedicated GPUs you'll find is discarded old junk in a cardboard junk box.
And then the bubble would collapse. Corps are already putting limits on token usage across the board because of costs. Increasing costs would significantly contract the hype bubble.
They're viable, you just have to think about them differently than how you think about support employees.
With humans, it's acceptable to have an "authenticate a customer" tool and a "reset the customer's password" tool as two separate applications. You can put in the manual that the latter can only be used after the former.
With agents, you can achieve the same outcome, but the constraint needs to be enforced by code, not job training and employee handbooks.
reply