Agreed. I think the stochastic parrot concept is useful to ground our expectations on LLMs for now, but it could outlive it's usefulness if there ends up being multiple jumps in sophistication similar to that of GPT-2 to 4 in the next 10 years.

If that happens, then stochastic parrot as an argument as to why a machine isn't thinking can be made pretty useless if One chooses to drag the argument further into philosophy.