Comment Re:Welcome our new overlords (Score 1) 62
That depends on what exactly you intend to do. Producing tensor models is going to be out of reach of most people simply because the training data you need is generally going to be quite vast, and it's a huge undertaking to get all of that. Then the hardware/time required to process all of that is even more onerous. But once the tensor model is built (trained) gaming GPUs are generally fast enough to generate content locally at an acceptable speed. They're a bit slower than dedicated ASIC (read: Tensor Processing Unit) and/or FPGA hardware that the big AI companies use, but far from impractical. And the better the GPUs get, or as in the case of Pixel phones, having a dedicated TPU, it will only get faster.
If you've ever done algebra and used a TI graphing calculator, you've probably seen e.g. the linear regression, exponential regression, sinusoidal regression, or other functions that spit out an algebraic function (i.e. "model") that you can then use it to extrapolate/interpolate additional data points around. This is the same basic principle around AI.
There are a lot of pre-trained models available on sites like huggingface that you can just download and use. Take this one for example:
https://huggingface.co/RomixER...
If you want it censored, you'll need to run it through another model trained to censor images. See https://huggingface.co/models and click on tasks. Toy around with these models enough, and you'll get a better understanding of how it works, and where the limitations lie. Chatgippity, and to a greater extent, AI hypists, more or less turn AI into this big enigma that it is not.
My understanding of openAI is they were supposed to be a nonprofit that would research, develop, and spit out models that they would make available to the public generally for free, but then changed to a for-profit model and stopped publishing them. That fact is why they're being sued by Elon Musk, who was one of the original donors, and if I were him I'd probably want my money back too if it just went into somebody else's pocket in the form of private equity shares. Either that or force them to make the models public again, but then the more recent investors would probably sue the company for tearing down its only moat.