Doesn’t generative AI need a whole other layer of technology to become reliable?
The AI needs to control some domain-specific model (like a poser skeleton for pictures of humans) that enforces the rules for how each modelled concept can actually behave, instead of trying to guess the output directly.