Knightshade
Well-Known Member
Not sure if you all follow Dr. know it all… But he’s a pretty sharp cookie. Multiple PhD‘s. Including physics and mathematics.
According to the math that he lays out in this video; FSD will be solved within the year. Couldn’t time stamp it; but if you start at 8:00 min mark you’ll get the nuts and bolts of it.
So…, two weeks is coming up quick!
That's a... weird... video... his first assumption is # of interventions per km, which he appears to base on... his personal driving in the car. So that's a...not great...assumption to apply fleet (and world) wide to start with.
He then tells us intervention rate will drop exponentially based ENTIRELY on collecting more data.... which is... overly reductive? Indeed, at 13:12 he admits he's just making an assumption that's true, and everything he concludes is based on that assumption.... which he thinks must be true because "that's how physics works"
I mean absolutely having tons of data is essential for training up NNs... but they're not the ONLY thing needed.
Tesla has already fundamentally changed their approach to FSD development at least twice since 2016, and changed the HW needed 1 or 2 times as well (depending if you count the 2->2.5 HW change). Clearly there's more to solving this than just:
Step 1. Collect data
Step 3. PROFIT!