Yes, thankfully the reasonable tech companies offering these services have decided to stop the training process after it was done once. The insane increase in energy consumption and hardware manufacturing for datacenter components and accelerators is purely coincidental and has nothing to do with demand for gimmicky generative AI services. Let's also conveniently ignore the increasing inference cost of more complex models, while we're at it.
Their goal isn't to replace a few staff members, it's to replace all of them, everywhere, across the globe. So they consider it a worthwhile investment. As to what we'll do when 5 people are in control of, manufacturer, and create literally everything?
Surprised to hear that my graphics card consumes enough electricity to power a small town for two months in the second it takes for it to generate an image. I'm getting incredibly good electricity rates if so.
I feel like most of the articles saying this are confusing the mining power usage (a constant load) with the transaction power usage. (Essentially nothing) Each transaction fee does incentivise more mining, but it's not a flat power cost per transaction.