We Have A Winner For Tesla Bull Comment of the Decade![/b]
Actually you don't. You need to read this op ed on Seekingalpha which talks about the potential path to FSD for Tesla and the implications of it in terms of ToboTaxi.
https://seekingalpha.com/article/4295418-tesla-autonomy-possibility-fast-progressIt is a very well researched and written article by someone who actually knows what they are talking about.
There are some key points that people need to grab hold of and understand.
1. Dataset size
2. The speed of computer learning as opposed to the speed of building the learning engine
3. The hardware in the vehicles required to manage all of this.
So what does the article say about these things.
Dataset size.
Tesla has approximately 600,000 vehicles with “full self-driving hardware”. These vehicles drive something in the ballpark of 20 million miles per day. A competitor like Waymo with roughly 0.1% as many vehicles can’t create a dataset of the same size.
and
Since Tesla is pulling from approximately 20 million miles of driving per day, it can rapidly build up very large datasets of different driving behaviours and collect new demonstrations to correct neural network errors. For a competitor like Waymo which might take two years to drive 20 million miles, this approach at this scale just isn’t possible.
The speed of the differing activities which will lead to a computer driven "product" which can replace humans for the same job (in this case playing computer games)
DeepMind spent about three years in research and development on StarCraft and then used imitation learning to train its agent to human-level performance in three days.
Playing a computer game requires incredible levels of interaction and skill/off the wall knowledge.
Three days, hold that thought.
Then we have the hardware in the vehicle. Now there is a point here which is not mentioned in the article because the author wants to make specific points. But I, from my own experience, am going to make another point.
Nvidia hardware was an evolutionary development in order to keep incrementally making car driving easier. But it was, in no way, fast enough or viable enough to take the leap from assisting to actually driving.
The Tesla hardware was a Revolutionary step which skipped several levels (years/decades), of evolution to get to the ability to actually drive in one single hop. Whilst using a fraction of the power of current systems (critical for EV's).
So
1.Deploy a new computer vision neural network that uses the new hardware’s 21x increase in video processing capability. We know Karpathy’s team is working on such a network, but we don’t know when it will be deployed.
Because to get quality data you need the compute power to decide when to get it.
So where does it lead the author?
The central idea here is: if Tesla’s approach is right, and if most of the manual work has already been done, then the steps to implementing it can be executed quickly, some of them at computer speed. Training the imitation learning networks, for instance, might take only a matter of days.
So why do I call the author the biggest bull?
On Investment opportunities, the author says
From an investment perspective, this creates an unusual and possibly unique (at least for a large-cap company) situation where the valuation logic for Tesla depends on a somewhat unknowable scientific/engineering factor that could rapidly change, causing the company’s rational valuation to jump 10x or more.
But finishes with
However, the underlying uncertainty is the feasibility of the technology: in particular, the near-to-medium-term potential for deep learning to match human competence in vision, behaviour prediction, and driving behaviour. While I can’t resolve that uncertainty, I will make two claims. First, if deep learning can master these problems in the near-to-medium term, Tesla will be the one to prove it. Second, if Tesla solves full autonomy, there is a realistic possibility that, from the outside, progress will appear blazingly fast, catching many people by surprise. We should think about technical progress on this problem as a combination of subproblems solved at human coding speed or human R&D speed and subproblems solved at data uploading speed or neural network training speed. A neural network that takes years to develop might take only a few days to train. If we expect progress to be steady, smooth, and incremental, then, from the outside, we might miss this process. We don't want to miss it!
Read it all. It is worth reading. It is worth reflecting on. Especially the robotaxi implications.
Also, you might want to consider the entire article in the light of the decision to deliver A/S before more advanced on road A/P.
In terms of A/P, the amount of miles is key, as it takes so many miles to generate so many anomalies and situations. Advanced Summon only needs "journeys" to create data. By putting the vehicle in the most demanding environment it is possible to think of, every single journey elicits extremely valid data. Half a million A/S journeys are worth hundreds of millions of miles in training data. The data contains pedestrian hazards, vehicle hazards, loose shopping cart hazards, high volumes of children out of control, vehicles in the most demanding visibility, backing out of spaces where they may not see other vehicles.
I expect extremely rapid development of A/S and FSD as a result of this single delivery of a feature.
Add to this the Agile delivery of the software (where you test, fail, fix and test again in a continuous loop) and you have the potential for Tesla to overtake all competitors in 6 months.
Advanced Summon, alone, will leapfrog the entire database of knowledge of all competitors, combined, in half a year. The learning software will consume it in hours. So long as owners continue to use it.