Comment by daxfohl

Partially related, is charging by token sustainable for LLM shops? If the compute requirements go up quadratically, doesn't that mean cost should as well?