Diffusion models aren't terribly power hungry compared to something like chatGPT*, but it's weird and honestly worthless idea in the first place, so please shit on it.
*Technically you can have a gpt text embedding to drive a diffusion model so they can be just as bad, but it isn't necessarily always the case.
I'm working on a theory, that the rise of livestreaming and the erosion of rights, or at least sympathy, for the disabled are connected. You can decide what kind of -posting this is, but the disabled have historically been shunned to venues of less-than-repute entertainment to make a living. Carnival workers, 'freak show' members, etc. And live streaming is just the next venue to pop off. As more disabled people, who are unable to work a regular job, give streaming a chance (www.twitch.tv/myelint tomorrow at 1 p.m. est) the expectation that the disabled will be able to make a good amount of money just playing video games is reinforced in society and the idea that we may need assistance, especially financial, becomes less accepted.
But even for the successful, it's just like working those older jobs. There is no advancement in the field of streaming, just as there was no advancement for working at a carnival. Even worse, creators on streaming platforms are not given the same workers' protections as a regular job, beholden to the 'gig economy' style of employment where you can have your job snatched out from under you and you get no attempt to bargain your own pay. Then there is the fact that you have to be constantly performing and hope you don't lose your audience.