/vt/ - Virtual Youtubers

>>75860983

>>75861410
>Surely it's not just the application of processing power.
As that other anon already explained it's pretty much that, basically model training is gonna be fucked until the bubble bursts. Right now no one wants to take the first step because you don't want to waste your cash, so you want to make sure you have good research and dataset before you start training, but you can't really be sure because testing it out is also really expensive, so you're back where you started.
I think things will get significantly better once the bubble bursts tho, the demand for computing power will drop significantly, the business will stop revolving around fishing for investors with bigger numbers, so smarter, more efficient base models will start to pop up, and if we're lucky devs will start supporting AMD properly so that people will have access to hardware that isn't horribly overpriced.