The core problem with large language models and the companies developing them is that they are an extractive business.

They would not be possible—and they will not be sustainable—without a constant supply of human-generated text, sound, and imagery to mine, but they are destroying the systems that have allowed for the generation of that raw material they depend on.