“@realGeorgeHotz @JackRhysider @comma_ai Tesla is absurdly efficient at inference, because we’re limited by a ~200W in-car computer and actually can’t brute force a solution with several kW of Nvidia GPUs”