3 ms·
> As I understand it, generative AI is a sort of average machine, in that if you ask it for a picture of a dog, it’ll give you the most average result for the p
by astrange 2y ago
> As I understand it, generative AI is a sort of average machine, in that if you ask it for a picture of a dog, it’ll give you the most average result for the prompt, model and whatever randomness there is in the system.
"Generative AI" isn't a single technology. Text and image generation don't actually work the same way (have the same model architecture), which frankly makes it absurd we invented them both at the same time.
But for text at least this isn't true; LLMs have fractal complexity, which is why they work. ("Average" text would just be the letter 'e'.) It's harder to say how diffusion models even work.
It makes them easier to handle if you train them to act like this though, but try playing with a "base model" and you'll see all kinds of inhuman results.
- pbhjpbhj 2y agoAIUI diffusion models of images are trained to make a noisy image into an image without noise (reverse diffusion). Then when an image is required, the system starts with white noise, and maybe a prompt, and then iteratively performs reverse diffusion - guided by the prompt - to produce a final image. You can enhance an image by [forward] diffusing it, then feeding it to the NN created to perform the reverse diffusion.
- astrange 2y agoThat's the idea behind how they work, but cold diffusion and latent consistency models show that you don't actually need any particular step.