And even if we did manage to limit/cull hardware development, what would keep people from continuing to find ever more efficient models runnable on contemporary hardware?
Bleeding edge models are putting the desk-job competencies of multiple professional fields into something with as many parameters as a single large rodent has synapses.
