Makes sense. Fortunately, scale is less of a problem for the autolabeler, so they can put a lot of processing power on this step. But yeah, they will need human

Makes sense. Fortunately, scale is less of a problem for the autolabeler, so they can put a lot of processing power on this step. But yeah, they will need human labelers for the edge cases where the model is unsure. I wish Tesla would publish their results so we could understand what they are doing and how much it is improving.