this post was submitted on 09 Dec 2025
109 points (99.1% liked)
Fuck AI
4811 readers
1381 users here now
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this makes me wonder what hoppens when the bubble does burst, because the tooling will stick around. when the next-next gen gpus all have this stuff optimised to shit as a matter of course (it's all vector processing, that's good for graphics too), and a studio can train an image generation model on only their own stuff as part of their normal rendering pipeline rather than spending extra energy, and there's actual qa for the results so the "look" isn't there, will there still be complaints?
part of me thinks there will.
I saw an "AI" tool recently which showed how you could create two very different poses for a character, and it would "tween" between the two in a realistic, convincing way. It could be described as "genAI" I suppose, but the company claimed they were very specific with how they trained the model and what it was intended to do.
There were still animators upset about it, and I get it. I'd probably be upset about it if I were in that profession. I'm certainly upset about LLM use in programming. But if I squint really hard, I can barely eke out a vision of limited, targeted, vetted tools which accomplish very specific aids to creators in their professional workflows.
That is not by and large how any of the services we regularly hear about are built and marketed. There's a wide gulf between ethically-sourced, limited professional workflow tools and the Copilots & Soras & Sunos of the world. I would say as a general rule, if something is produced based on a "prompt", it should be immediately viewed with immense suspicion.
i saw a tool like that in like 1999. like, not super-realistic but convincing enough. above animorphs level at least.
idk about immediately throwing out prompts either. i did a big half-rant about this the other day but the current gen of models are all vector spaces. they are basically multidimensional topographical maps with the prompt as the starting coordinates and some traversal algorithm as the means of producing output. as long as they stick around the prompts are needed.
Beyond that, there's definite value in being able to create queries using natural language, and not needing to always know the specific technical terminology of something while still being able to get pointed in the right direction.
It's the whole non-deterministically regurgitate a poor quality combination of all your stolen training data while not citing sources and using absurd resources that I have a problem with, personally.