While that may well apply to the overall AI program as such, I don't suspect it is descriptive of LLM.

Throwing more parameters at LLMs is no longer yielding appreciable improvements, and once you have thrown all the data at them you can find, that dimension is done also.