>>99940293Wonder how you will be reacting when they make AI models that actually can intuit about things they don't understand. Trick question- they did.
The only reason it isn't done well is it is obscenely expensive to create a layer above the AI models that moderate their outputs in that way, so the AI will always try to do a best fit because that's how it was trained to do it, it was never trained reliably say "sorry I don't understand that", it was literally trained to be as useful as possible.
That shit costs several hundred thousand to do at a competent level, that ain't profitable in any sense, that'd kill the company who tried to do it in mere months.
Instead they will just give multiple outputs and hope one hits the mark.
The AI can never intuit about this because you never gave them the context needed.
They could be an idiot and typing "tell me about apple" could be them meaning "tell me about apples" or "tell me about Apple".
This is a very simple concept, but there's a tillion other subtle scenarios like this and it's also the reason it is easy to jailbreak the AI models because of how nuanced language is, it's not easy to have an AI model be able to discuss, say, sex, and censor RP sex, it just ain't possible without models that actually think about their inputs properly, and not the hacky CoT shit they do now. In fact, just like I said on /g/ years back in the CAI era, I knew them adding CoT abilities to AI would make it easier to jailbreak them, and they are in fact easier to jailbreak.
>>99940390And if the AI training assets were paid for it would suddenly change things?
Or will you still screech about it?
Going back to the original point of this thread, the fact that the AI dress was then used to make human works, does the fact the dress is AI suddenly invalidate the drawn artworks just because it was inspired by the AI, even if the AI training was based on training on unlicensed works?
