Raw LLM Responses

@9:07 Using AI for reference: I think this is something it can be good for, but not for details like you said a reference usually would be for. I've used AI references in the past for pose/composition reference, but then fill a reference board with real pictures of the details that need to go in. Granted, I also work in 3D and don't really copy references that close (more to just "get the broad strokes"), so crooked toed lion with no ears wouldn't exactly be a problem. That said, I've used an AI reference as the primary reference exactly once in the dozens of 3D models I've done - a few pieces I've had an AI image somewhere in the reference board for vibes or small section of it being exactly the way I wanted. The AI bubble is starting to collapse. The return of investment hasn't materialized, and it's feeling like a mad scramble to push it out to absolutely everything to force more users into the ecosystem, so they can justify the insane costs of training the models. I use Large Language Models for work almost daily: They work like a personal intern, doing some busywork so I can save a few minutes from writing an SQL prompt, or a powershell script. For SQL the stuff works pretty well, with powershell already I'll be fixing and tweaking the results to get things working. These tools are in their infancy still, and the output is questionable at best. It'll be a few more years before they're any real threats to jobs anywhere, and it's the same as with image generation: Unless the next generation tech in image and language generation gets an order of magnitude cheaper to train, and can do this training with an order of magnitude less training data, AND can do a better job than the current models can: They are simply fucked.