@0c1a9324 
What, are you saying O(n^3) complexity for what amounts to a bullshit generator for mansplaining as a service turns out to have limited demand? 

 @0c1a9324 
People are finding out it really can't do what it's promoted for. (For specific purposes, it can be very good. But for general uses, nope. 

 @0c1a9324 Depends on the type of neural network. Generative AI as smart assistants in music and graphics software is becoming more widespread, but there are small models which often run on the local GPU. So are upscaling AIs in games--mostly proprietary and owned by Nvidia, they allow for extremely realistic visuals, since the graphics engine only needs to render the scene with all the raytracing, radiosity, subsurface scattering, volumetric fog, etc., at a low resolution and low frame rate, using less than half of the GPU, while a huge chunk of the rest of the GPU upscales the lorez images to 2k or even 4k and generates intermediate frames. 
It is only the big LLMs that are quite useless right now, and they're too expensive in terms of energy and resource usage to be used as a toy. Smaller language models running on local GPUs will probably become a common thing in games very soon, as they allow NPCs to talk in a more natural way and allow for unscripted dialogues. 

 @0c1a9324 I am constantly pointing out at work that AI is an umbrella term. Just educating people that AI does not mean chat GPT or other GPTs is exhausting. NLP is not GPT. I know how Sistphus feels some days.