Good points which make the case for open AI to always be on par with the best closed models, anything else would be an existential hazard indeed.