-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Sir Guacamolaf
The good kind of fat
This brings to mind an analogy. My sister got pregnant last year and the doctor told her that delivery would be MAYBE Jan 15, and definitely before Feb 1. Actually delivery was quite later on Feb 12 and there were lot of complications. Should she sue the doctor?
PUSH ELON! PUSH!! I CAN SEE THE FRONT NOSE CONE!! PUSH!!
I think AP2 and FSD is a complete farce, fabricated to get rich AP1 car owners to trade up to an AP2 car. AP1 is amazing, and some day FSD will be even more amazing, but it won't be for a very, very long time. EAP will do very little more than AP1 even at its peak, and it certainly does not justify the huge cost increase or the depreciation hit it will take to go from AP1 to AP2.
JeffK
Well-Known Member
There are some videos that proves you wrong. With enough training FSD can work just fine as they demonstrated, unless you think the videos were somehow faked...
That video was on a specific trained route with hundreds of runs. Try that same software on an unknown route and the car would not make it to its destination.
We are years, and years away from FSD being a reality from both a ready and a legal standpoint. Ive seen the videos, and it does not change my stance one bit. Are there scenarios where the car will be able to FSD, sure, will it be ready for prime time and legal, within the next few years, absolutely not.
JeffK
Well-Known Member
And yet that's precisely how neural networks work and through generalization can tackle unseen routes. The amount of data needed for that is enormous. Every company currently pursuing autonomous driving is using neural networks. Tesla is in the best position for data gathering.
Thank you for your informed opinion.
You are so wrong that even Sterling Anderson, the Director of Tesla Autopilot Programs disagrees with you.
(actually 21 mins)
MIT Technology Review Events Videos - Delivering on the Promise of Autonomous Vehicles
But i don't blame you, being clueless of how something works leads to buzzword theories.
neural networks are a buzzword to you.
JeffK
Well-Known Member
Last time I check he doesn't work there anymore.
Here's what Elon says on the subject
Elon Musk on Twitter
I don't blame you, you're from Michigan (where there are a bunch of Tesla haters)
I've programmed (built and trained) DNNs myself and am very familiar, but thanks. MobilEye's NNs were used for AP 1 as well. In your video Sterling mentions the importance of NNs, the need for gathering large amounts of data, and the need to fine tune the algorithms for corner cases... Every supplier of autonomous vehicle technology does this currently. Did you have a question?
Last edited:
Last time I check he doesn't work there anymore.
Here's what Elon says on the subject
Elon Musk on Twitter
I don't blame you, you're from Michigan (where there are a bunch of Tesla haters)
The fact that you understood 0% of what the director of Autopilot said, including his statement in 18 minutes of how easy it is to do a demo which is exactly what he did and your reponse is that he doesn't work there anymore and a tweet from elon shows how naive you are.
I would look to have a live chat with logs with you so you share with me what you know about DNN.
JeffK
Well-Known Member
correction: Former director of Autopilot
in other news:
Tesla has settled its lawsuit against self-driving startup Aurora
Who let this guy sit at the grown-ups table?
JeffK
Well-Known Member
Let me know if you can find a single maker of autonomous driving tech that doesn't use neural networks and loads of data and simulation.
Of course it's easy to make a demo, but as I previously mentioned you need loads of data and that needs to be collected by drivers in various driving conditions. This takes time, training networks takes time, adjusting corner cases take time, validation takes time, regulatory agencies take time to review data, etc. The demo never meant it was ready to go and the major data collection couldn't have even really started until AP2 cars were on the streets. For someone to be frustrated with the FSD timeline, it means they had unrealistic estimates. It's going to be at least a year minimum.
Elon's tweet mentioned it could be June before you see FSD start to diverge from EAP. It's still April and even then it doesn't mean full self driving, it means maybe a feature or two different from EAP.
Since JeffK claimed he knowledgable about neural networks and have "programmed (built and trained)" and is "very familiar", would love him to tell me what he knows and lets build one live and i will supply him with the data he needs. This happens live with no forum delay and running to google and wiki for help.
If he can't handle that then doesn't know jack.
Literrally i can ask him one or two question and he goes crickets would really expose his knowledge.
Obviously you can't do it through forum cause he simple google cheats.
JeffK
Well-Known Member
TMC is not the place and I'm not going to resort to snippiness, but if you'd like to learn the very basics of neural networks and AI I can point you to Andrew Ng's machine learning coursera course and the course from Geoffrey Hinton for starters. Nvidia also has some great tutorials on their DIGITS software using Caffe.
Similar threads
