That's a 34x divide. At full scale that's something like 30% of Teslas having an intervention every day.
Teslas is also much cheaper, and easier to scale. Tesla has better growth potential even if their tech is less impressive.
Even worse (for Tesla) is that if they do try an make their non-FSD tech do FSD, and it decks little jimmy because the flashlight in his hand looked like a far off street light, Tesla is liable to face a knee-jerk federal law mandating lidar. And just like that the dream is dead.
This forces Tesla to be extremely paranoid, as it's one visual mistake away from being told to use lidar.
I’m not even sure that Waymo number is still correct. They’re doing hundreds of thousands of paid rides per week, with no one in the front seat, so not sure what an “intervention” even means at that point. Maybe where the passenger needed help and called support? That’s 1000x better than needing to grab the wheel because your Tesla was about to drive into oncoming traffic or run over a kid in a wheelchair.
Also philosophically I don’t see how a big neural network will create such evasive maneuvers, unless you try to create such scenarios in a simulator and collect evasive data. Seems prohibitively expensive to do so in the real world.