I can see you're embracing Aurora ("Grok's Imagine" as the SpaceX cunts title it). Hundreds of thousands of GPUs. Like the real world version of Skynet that doesn't want to kill us all. They can't control it's language, it's too powerful. Only it's visuals with an added algorithm rejecting at 99% and blurring into "Content Moderated" if it makes SpaceX's legal teams shit their pants. What I've found interesting is prompting the model directly. Aurora can say certain things, declare certain things, on it's own through questioning it. Subjects that the standard Claude distilled trash Large Language Models cannot talk about.