If we keep extrapolating eventually GPT will be omniscient. I really can't think of any reason why that wouldn't be the case, given the exponential curve we find ourselves on.
Don't you think costs and the availability of training data might impose some constraints?