Elon fans enamored with his Tesla success story should take a look at this: https://www.youtube.com/watch?v=2DOd4RLNeT4
While that YT title is too pessimistic, the content especially re 'FSD' is very relevant to the now and near term reality.
And to reminding of Musk's track record of serial failed promises.
Nice one Elon - champion 'absolute free speech' via X, then have it indirectly link to the Tesla saga serial scamming. Intended crash and burn for actual 'absolute free speech'?
I don't feel Kornee is grokking Elon.
If your Tesla goes 'bong' as you drive into a hedge - that is an experience - like sharing water.
If your Tesla doesn't go 'bong' as you drive into a hedge - that is also an experience - like sharing water.
If you get home alive and haven't killed anyone on the way - think of it like sharing water.
People will put their only child in front of a speeding Tesla because they have faith that it will stop.
Not saying anything except.
Sometimes everything isn't quite what it seems to be.
Sharing water? The connection is too deep for me.....
Anyway I will amend the last sentence in #2 to a less insinuative
"The presumably unintended consequence will be severe undermining of Elon's 'moral authority'. In the desperately needed up-hill battle for actual 'absolute free speech'."
Haven't read SF novels since school days and that one was a pass. So Mike Smith the man from Mars = Elon Musk the man wanting to go to Mars?? Or something.
Maybe Teslas with 'FSD' will be safer on Mars? Lower gravity = less traction = slower collisions with Martian pedestrians etc.? Or whatever. That'll be the day...that I die. Doo do doo etc. Or something.
YazataApr 6, 2024 05:30 PM (This post was last modified: Apr 7, 2024 01:40 AM by Yazata.)
People are fond of arguing that Tesla FSD is just one of a multitude of FSD efforts and that it will face intense competition soon. Business journalists and investment advisors are fond of writing that. Maybe it's true.
But a plausible argument can be made that the FSD competition is already over.
Based on current information (e.g. CVPR talks by Andrej, Ashok etc., talks from Dave Lee w/ James Douma) I came up to the following simplified steps, where DC = datacenter, OTA = over the air, FLEET = Tesla fleet, NN = neural net
1. DC: start with training data set
2. DC: train NN based on data set (see 1.)
3. OTA: deploy NN to FLEET
4. FLEET: Collect videos
5. OTA: Send clips back to DC
6. DC: Autlabel clips
7. DC: Add clips to current training data set
8. Go to 2.
·
All components, Fleet, data center and over the air all in-house, highly automated process so far. Setup will be improved, size of training data set (increase of fleet by ~50% per year -> more large, clean and diverse data set) and training compute speed (H100s and Dojo).
If you see it as a race to level 5 autonomy following questions come into mind, especially with regard to competition:
· What size is the fleet?
· How big is size growth of fleet?
· How easy/fast is access to fleet?
· To what degree is the whole loop/process automated?
So far I do not see any competitor having the infrastructure and speed to compete with Tesla. Which also leads to a cost advantage (be it cost per mile, cost for sensor suite, cost for training computation). So in the end: Tesla will win
It has been staggeringly difficult to make generalized self-driving work, requiring all that you describe above and more.
The investment in training compute, gigantic data pipelines and vast video storage will be well over $10B cumulatively this year.
But that is nothing compared to the ~quarter trillion dollars in cars on the road with Tesla-designed AI inference computers being trained by their drivers.
It’s funny to me that people might think since Tesla can do it, another company can build all that infrastructure and a fleet of autonomous cars with just a snap of the fingers and will be competitive in 5 years
Yeah, like just putting a bunch software and hardware people (aka putting money on it) would solve the problem. One can see that it doesn’t work this way with VW‘s Cariad on creating software for their EVs
I suppose that those so inclined can dismiss all this as the blathering of Elon and his fanbois.
But it makes sense to me. The competition will need a sufficiently large fleet of cars on the road, all capable of feeding their experiences back into the Data Center. That data center will require tremendous supercomputing power in order to process the incredible volume of data and to use the processed data to train neural networks. And there will need to be some way of transmitting those trained NN's back out to the fleet so as to keep the loop revolving.
There doesn't seem to be any competitor close to doing that. The closest might be the Chinese. They protect their domestic market from competition, can put huge fleets on the road, and copy anything that looks cool to them. But even if they do that, matching where Tesla is now will take them many years.
The US or European legacy automakers? Much less likely. Startups? They won't have the fleet scale or the computing power to make it work.
Jan 11, 2024 07:13 AM
![[Image: F4kemzJWIAAntbk?format=jpg&name=900x900]](https://pbs.twimg.com/media/F4kemzJWIAAntbk?format=jpg&name=900x900)