I think what we have is a toy and will remain a toy, just like Eliza was 60 years ago. Academically fascinating, and given the constraints of the era, genuinely remarkable, but still a long way from really being useful.
I'm already getting bored of seeing 95% amazing 5% wtf AI generated images, I can't fathom how anyone else remains excited about this stuff so long. My slack is filled with impressive-but-not-quite-right images of all sorts of outrageous scenarios.
But that's the catch. These diffusion models are stuck creating wacky or surreal images because those contexts are essential allowing you to easily ignore how much these generates miss the mark.
Synthetic AI media won't even been as disruptive as photoshop, let alone the creation of written language.