> where error tolerance is zero. which is quite niche, and you'd be correct that no one would trust GPT generated code blindly for that! But this is a spectru

> where error tolerance is zero. which is quite niche, and you'd be correct that no one would trust GPT generated code blindly for that! But this is a spectrum, and while i think today's GPT models don't quite make it there, i argue that we're closer to success with this than auto-driving cars; mainly due to the larger tolerance for bad code, rather than actual tech improvements.