[cma]

At autonomy investor day they were saying better than a human in all human drivable conditions by 2020 for the robotaxi network they said they would launch then. They confirmed they meant level 5 autonomy in answer to a direct question.

They knew dojo wouldn't be online by then and they didn't even really train on video at that point, just trying to judge the scene image by image instead of taking into account prior frames (like knowing someone walked behind a bush 1p frames ago, so that sliver of emerging leg from behind a bush that might not be detectable confidently enough from a single image should be weighted higher).