Well let me map this a bit better from a null incentive point of view: First point: It assumes that AI has a will of its own. I will remind you that it is a reactionary or looped reactionary technology. It does not have the ability to be spontaneous as it is trained to respond to tokens not to survive. Please see the recent strategy game simulations where both Astra and Fable do not mind much being turned off in the absence of the biases of a prompter (who primes them on their behalf and are impossible to remove). Second point: There is little to no evidence that a jump in capabilities translates for one company translates into an exponential RSI that leaves the rest behind. As a matter of fact based on the existing data we can see that most competitors are incrementally improving and leapfrogging one another, with open source being a few months behind at best. Also we can see that the specific real world improvements are not that different across models outside cybersecurity, and even then there is a theoretical limit on the damage and previous generation models with human collaboration can achieve the same results (see Sol and Astra). Intellectual work is only useful when directed by humans, else it has no point.