Raw LLM Responses

I've worked on AI. It's fascinating and it absolutely is the future, but I absolutely despise LLMs. AI should be narrow and fit for purpose. It should be trained only on relevant data. Stuff I'm talking about here is local weather prediction,x-ray scanners that don't fall for doctor's biases, melanoma recognition that outperforms doctors, scanning old structures with a drone to look for weak points to reinforce. These are all AI I've seen or read about and I absolutely want in this world. And the same goes for image generation. In some cases with little or poor quality data, a small AI will actually get better when partially generating the data set. I've once seen a model that was best and most precise when 80% of its data was 'fake'. The reason this worked is the exact same reason why Nightshade works. It generates images on the patterns that it's learning, but the added randomness makes it so that it's not overtraining on the generated images. Nightshade doesn't reinforce these patterns but it disrupt them. It claws at image generators fundamentally. You just cannot fix this unless you can remove the nightshade from the image first. 100% of the generated images most people see come from wide image generators trained on other people's work, and I hate even more than LLMs because words are a lot easier to shuffle than than images, and seeing the bastardised traces of the beautiful art it's trained on is soul crushing. If you just get even a little too specific with your prompt without specifying an artist, it can already pigeonhole itself into a certain artist's style. The image generators got me rooting for Disney, and I'll never forgive them for that. And disabled artists bring such a huge variety to art. Something that can only happen through their struggle and lived experience, and it enriches all of us and society as a whole. In my opinion, the most impactful art is done by those who are disabled or traumatised. Image generators can't get even close to a fraction of this. Every line in a pained person's work tells a bigger story than anything AI generated ever could. It could write a work the length of The Subspace Emissary's World Conquest, and still not have the combined soul, meaning and expression of the opening line of 1984.