James Douma is an AI expert and he explains how Tesla FSD Version 12 is vastly better than Tesla FSD V11. Douma predicts that FSD version 12 will be able to have a 100X reduction in human interventions compared to version 11. This should be in the 3 month timeframe.
Tesla FSD v12 has eliminated the hard coded rules and replaced it with pure neural net AI. This enables Tesla to train and improve the system by finding video good drivers handling situations. The system then learns from the good examples.
V12 is going to be over 100x reduction in interventions compared to V11. This is not an incremental upgrade, it's a leap forward.
Thanks @Rebellionaire3 for giving me a chance to experience V12 path planning and control. It's impressively better than V11 and I predict that most… https://t.co/IHjdTs5KQJ
— jimmah (@jamesdouma) March 5, 2024
There are going to be some serious regressions at first – I didn't experience them but I'm confident they are in there. But I expect they can be polished out and will leave V12 far ahead of V11. Most interventions are for trivial things and those will mostly go away.
— jimmah (@jamesdouma) March 6, 2024
It's based on a small sample, but the sample is far out of distribution for V11 behavior so I'm comfortable with it. 10x is way too low and the sample is too small to support 1000x. So 100x plus is the right ballpark.
— jimmah (@jamesdouma) March 6, 2024
Thanks, that is an accurate assessment.
Training compute is currently our limiting factor, but that issue is being resolved fast.
— Elon Musk (@elonmusk) March 6, 2024
Improving FSD
I've been asked a few times about how and whether FSD can continue to get better without improvements to the hardware. This question is often asked in the context of a misunderstanding of how neural network optimization and improvement generally works, so I'm going…
— jimmah (@jamesdouma) March 3, 2024
Hardware 3 in Older Cars Is Still Enough to Handle the Improved FSD AI
FSD running on HW3 today is almost certainly orders of magnitude better (faster / smaller / more performant) than the first versions that came out when the platform debuted. And it will continue to get dramatically better for as long as Tesla cares to continue investing developer resources in making it better. There is certainly a point at which it’s cheaper to upgrade the millions of cars on the road than to invest the development cost needed to 10x the platform performance. But that day is not today and will probably not come for some time.
So why does HW4 exist if HW3 is adequate? Because silicon continues to get better and cheaper. After a few years it’s actually cheaper to move to a newer, better device than to continue using the old one.
The rate of methodological improvement is so fast it can be hard to believe. To give a sense of it: for a given level of performance we are seeing more than a 10x reduction in the hardware requirement for each year that passes. Better libraries, compilers, frameworks, and automated optimizers are part of this story but additionally new methods of quantization and of distillation, new architectural innovations, new and better data curation methods, larger datasets of higher quality data, new methods for automating hyper-parameter search and even of gradient descent itself are discovered regularly. So many powerful new methods have been uncovered in the last 12 months that everything running today will be 10x faster in a year even if we don’t find anything new. But we will find new improvements because we have every…single…year for the last decade.
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.
Yet still officially level 2 self-driving. Either people are over-exaggerating progress, or the regulators are dumb.
Yeah, and still at what I’ll call Insurance Level 1 too, where it’s just traditional insurance, blaming the driver mostly, and not the car, whenever there’s an accident. Insurance Level 2 would be FSD 4 or 5 + Blaming the car – i.e. manufacturer – or external circumstances, and data from the car contained in a black box, or maybe simulcast to outside storage to tell the difference. Neither manufacturers, nor in most cases, anyone responsible for external circumstances, will be willing to accept that kind of legal/monetary responsibility, so Level II may never happen. Insurance Level III could be Level II + some new kind of no-fault insurance in all circumstances. Essentially, this puts some private/public insurance authority in charge of determining fault, and paying out regardless, in fixed pre-determined amounts. To get to Level III would require major social, legal, and private sector cooperation.
Level IV would be accident-free insurance safety requirements. There will never be any EV to achieve Level IV.
In such a litigious society, can you blame them from not wanting to cross the threshold until they are absolutely sure?