Tesla Autonomy Investor Day described their software.
Tesla has a shadow mode where they can compare alternative software algorithms against the current software. Tesla will only proceed from shadow mode to early adopter software rollout if the Shadow Mode software proves it is safer .
Tesla now has over 70 million miles of fully engaged Autopilot driving.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
Very good point! And that may be why a human brain and sensory/motor system can’t perform driving tasks until they’ve reached a certain level of development.
In the future we may have robot operating systems that come pre-trained with motor control, object permanence, spatial integration, etc. functionality. Then any particular task (like driving) will only require a small amount of additional specialized training data.
the problem is I don’t think that is quite true, the statement “no human drivers needs to drive millions of miles to learn to drive safely”. the problem in that is it ignores connectionism. To drive, a human needs to learn motor control, learn spatial integration, learn object permanence, learn the ability to perform sequential actions, learn language, learn reasoning, etc. etc. etc. The driving is just the ‘cherry on top’ so to speak.
If you look at all the work the human brain needs to accomplish before it gets to the point where it can drive, I’d bet that that is the equivalent of ‘driving millions of miles to learn safety’.
Just a comment. Something is fundamentally wrong with the DNN of today. No human drivers needs to drive millions of miles to learn to drive safely, and NVIDIA and (implicitly) Tesla are talking about hundreds of millions to billions of miles. To get sufficient data to train their networks..
So something is really wrong. Is it the architecture? Is it the learning functions? Is the error calculations? Is it back-propagation? Who knows…
I foresee that someone “cracks” the question of getting a more human like learning profile, where a few hundred miles would be sufficient, and then this inventor will blow away the competition.
..Gathering data is outdated. When it comes to self driving it becomes clearer everyday that gathering data and using AI is not enough. That to build a good model you need to add a huge amount of logic the old way. This is the best understanding at this point.
Andrej Karpathy in a video about software 2.0 stack, explained that he spends 75% of his time curating the dataset and only 25% on the models and algorithms. (8:45) https://www.youtube.com/watch?v=y57wwucbXR8
That would require a smaller team. They don’t have to pay a group of people to chaperone hundreds of cars like Waymo does. Tesla is in the red partly because they are investing large sums into engineering and development. Elon also knows how to find the best people, does a lot of the hiring interviews himself. Then he get’s them to work harder and longer hours.
New applications or innovations often comes from small teams or companies. Maybe in the big companies there are more group dynamics at play, if you have someone with a weird idea or a viewpoint that is dramatically opposed to the majority then it’s easy to shoot down or ignore. Elon is keeping the hierarchy relatively flat (but he can fire you in a heartbeat) and because he is a smart engineer and actively involved he can make the right calls. Dumping LIDAR and focusing on creating large datasets trough realworld data is certainly unique in the industry.
Other then being the first to market with a massive installed base. Once you have that it gives you some advantages. The risk of course is some other company springboarding from your path, and outdoing you. But that installed base advantage is massive.
There are multiple paths to owning a market, But I dont see those capable competitors being able to so outdo Tesla in other areas that they will be able to push past Teslas advantage of effectively turning on hundreds of thousands of self driving vehicles in a short period.
Jepp, think you are right. Some chinese car manufacturers have *all* the data of their users…
How do they get the money for this? How can they afford to hire a critical mass of AI-engineers to get this done at the same time as the company is still in red?
Competition is nothing but good from a consumers POV. Kudos to the Tesla team on engineering their Neural Network Computer, currently in every new Tesla, and the team is working on Gen 2 NNC! The data will speak volumes, and will be difficult for safety regulators to ignore.
Good luck, Vimeo is way ahead of the pack, there are other, very capable competitors, they are all working very hard and there is no reason to believe that Tesla will become the safest.