Wonder if LLMs in autoreasearch loops would be able to complete tasks geohot has in mind in say 100x average token budget.

If the answer is yes, the argument doesn’t matter: you just run the loop and wait for llm analog of moore’s law to get costs down.