Maybe, but aren't LLM companies burning cash? The efficiency gains I see from LLMs typically come from agents which perform circular prompts on themselves until they reach some desired outcome (or give up until a human can prod them along).

It seems like we'll need to generate a lot more power to support these efficiency gains at scale, and unless that is coming from renewables (and even if it is) that cost may outweigh the gains for a long time.

I would say anyone who sees that happening is in denial, because all the proof out there points in the opposite direction.

There are also no indications that it's coming. Anyone who assumes so is delusional.