My former colleague Asa Dotzler tried asking generative AI models to create pictures of people writing, playing an instrument or playing tennis with their left hand and it went exactly as you would imagine it to go.
@GhostOnTheHalfShell@PixelProphecy That looks kind of like programming, though. Surely they should just shovel garbage into a data centre full of ASICs until the output magically improves.
looking at the bind bending progress in computers in the last 50 years
I am astonished that anyone would think that in 10 or 14 years that AI will look anything at all like it looks today
but why let reality get in the way of fun ?
I mean, can you imagine if S Jobs had let the conventional wisdom about say GUIs influence him, where we would be today ?
@failedLyndonLaRouchite 10 to 14 years from now we'll be battling with the devastating effects of global warming and nobody will give a **** about a poor quality, copyright-infringing stock photo generator
@gabrielesvelto object or concept placement within the image is generally broken with practically all of them (because of lack of caption quality - anything you want a model to see you have to specify in the training captions). Stable Cascade which got released 2 days ago might have a higher success rate, although here I'm not sure either. I will report back.
Add comment