Well, I'm like an informed user instead of a leading expert, but there is a very big chance that generative models might crash and burn. The architecture, either transformers or diffusers, is hard to control (hence hallucinations). That's bad for critical tasks, so it does not sell well...

... in many industries. Mediocre "art" and botched writing does not make a lot of money, and the promise of "AGI" will not keep investors hypped forever. Drug discovery might be a good use, but testing and synthesis is still a thing. Energy (oil exploration, cargo automation, fusion control) is...