- This topic has 19 replies, 9 voices, and was last updated 1 month ago by
.
-
Posts
-
Yeah agree - if AI is used to make the face look more realistic on a HAIR morph or something, fine. But not if its the actual FACE that's being AI'd.
More slop being requested - I have blocked about 6 different creators (probably all the same guy) from renderhub for selling this utter pish.
I really don't like using AI at all now, but when I was still trying to use it for memes, my understanding of checkpoints and LoRAs was that they were the strict wranglers that forced Stable Diffusion and all derivatives to focus on a very specific set of instructions, instead of it making total slop that the user had to play guessing games with in order to produce anything believable. And that seems to be the only way AI does shine, is when it is so restricted that it's not an everything model. But the people at the top dislike that because they want it to generate your 3d model, as well as snitch on you. Which is why the files for Diffusion have become so overblown you need a datacenter to host it instead of your local PC.
Even IT proponents of a type of AI that serves a logical function, have been harping on forcing it to have a more narrow specific focus, such as troubleshooting network problems. It doesn't need to be good at everything, just the troubleshooting of the network. But the tech bros have been deliberately undermining this school of thought by making them everything models that just become more gigantic for no good reason. They are creating a problem of the base model itself being so overblown that it requires the datacenters they want.
And the corporations who graft it into their software are only too happy to jump on the bandwagon, because they get financial incentive we never see from it. Many companies feel like they have no choice but to participate in the bubble for their own financial survival. That's largely an indication of their leadership being less tech savvy than those they employ, who have lived, breathed, and eaten tech as a matter of course for the last 25 years. It's the difference of looking at it solely from a perspective of financial gain on the rigged market, rather than tailoring it for a specific function. When you're an IT guru who's job it is to make sure things are secure and restricted, and you have to contend with the short sighted bosses who are fine with AI opening vulnerabilities as long as they get kickbacks, there's going to be a conflict of interest there.
So my read on it is that checkpoints and LoRAs were the answer to this disparity in philosophy that force the AI into something useful, with a narrow focus, instead of just throwing spaghetti at the wall and hoping it did something useful. It's the utter lack of restricting AI that is the problem. And it shows with how sloppy AI is when you look at it through the lens of surveillance, how it's been recklessly used to jail the wrong people repeatedly. The tech bros want it to be an everything model. Restricting it is not the issue. Letting it run wild so it can devour everything to make itself smarter is the problem. They are literally trying to create the Master Control Program.
I'm not anti Ai far from it, I dont like these vendors scamming folk into buying their shitty models that are misrepresented!
I would probably be ok with AI if it was highly restricted, focused, and had meaningful gaurdrails. But because they want it to become an omniscient, omnipresent, omnipotent self aware program that has the power over life or death, based on intelligence that has a complete inability to understand nuance, with none of the safeguards in place, I have huge problems with it, and how they are inserting that everything model into, well, everything. Because they do not intend even a little bit to use it for any kind of good. They pretend to, but the body language is far different than the sales pitch.
- You must be logged in to reply to this topic.
