OT more FSD discussion:
I am very bullish that if Tesla can achieve accurate FSD, it will be very bullish on share price.
I am not confident at all that this is happening soon. Frankly, I have no idea how anyone here can have high confidence, unless you work in computer vision and are familiar with this specific type of problem.
As background, I work in building machine learning / deep learning algorithms that end up running on memory constrained systems.
I was always skeptical about FSD on HW2 in 2016 because of computational limitations. Images weren't even proccessed at full resolution - they were downsampled. Not a good start.
Before HW3 came out,
@jimmy_d was excited because some big neural net was uncovered that was used for all cameras and took not one, but 2 full scale images as input.
Exciting yes, but 2 full sized images is a far cry from 6 seconds of video (if say that is what Dojo is going to be training on).
My point being, I think we should be excited that Tesla has the data and possibly getting the computational resources to train a great FSD net. But I don't see any evidence that it would work on HW3.
I am curious about peoples thoughts on why they think differently, other than "Elon said so"