Yes, that’s… the complaint. Umpteen hojillion dollars have been spent turning LLMs into a dancing bear, with nobody seeming to ask for a better text-to-text approach.

Meanwhile image and video shit keeps going crazy, just by removing noise.