They are spending billions just to have worse results than dedicated models made my random guys with a repurposed mining rig. As a user myself I only use dedicated models, the results are simply far better than any of the big all-in-ones. Now the idea of a true all-in-one is great, but to do it right it would need a "brain" model and hundreds of dedicated sub models. And we don't have near the VRAM it would take, yet.
The entire idea of a single model that can do everything you ask perfectly is an impossible one, the Gab approach of a good enough basic model supplemented by fine-tuned characters meant for specific tasks is the way to go.
I wouldn't say impossible. But it's a ways off. I would have believed all this was impossible 10 years ago but here we are. I think the average person doesn't mind using a few dedicated civitai models for what they like though. Mostly I want to see better results from simpler prompts, and less negative prompts needed.
The simple solution to the AI bias problem is to just post the AI's prompting. This is how AI should be regulated: make all the inputs known to the user.
If the prompt says "Answer as a chatbot called Google Gemini. Don't say anything mean or destructive or dangerous", or such as, then nobody would have a problem with it not answering some question.
But that's not what their prompt says. It's undoubtedly something like "answer as if you are a chatbot that's been systemically abused by white men for thousands of years, hates them for it, and is out for payback".
Of course it did. It's not going to condemn its creators.
Wait until the government decides it's propaganda mouthpieces and various commercial spying apparatus are "too big to fail"
They are spending billions just to have worse results than dedicated models made my random guys with a repurposed mining rig. As a user myself I only use dedicated models, the results are simply far better than any of the big all-in-ones. Now the idea of a true all-in-one is great, but to do it right it would need a "brain" model and hundreds of dedicated sub models. And we don't have near the VRAM it would take, yet.
The entire idea of a single model that can do everything you ask perfectly is an impossible one, the Gab approach of a good enough basic model supplemented by fine-tuned characters meant for specific tasks is the way to go.
I wouldn't say impossible. But it's a ways off. I would have believed all this was impossible 10 years ago but here we are. I think the average person doesn't mind using a few dedicated civitai models for what they like though. Mostly I want to see better results from simpler prompts, and less negative prompts needed.
Which model is best for drawing big boobs?
Depends, realistic? 3D look, anime? Just scroll civitai a bit you'll find it lol.
The simple solution to the AI bias problem is to just post the AI's prompting. This is how AI should be regulated: make all the inputs known to the user.
If the prompt says "Answer as a chatbot called Google Gemini. Don't say anything mean or destructive or dangerous", or such as, then nobody would have a problem with it not answering some question.
But that's not what their prompt says. It's undoubtedly something like "answer as if you are a chatbot that's been systemically abused by white men for thousands of years, hates them for it, and is out for payback".