diplomat33
Average guy who loves autonomous vehicles
Elon is using the hackathon to find new talent for the FSD team. If Tesla finds some smart new talent that can contribute to the development of FSD, that will be a good thing.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon: "Feature complete for full self driving this year"
- Thread starter dckiwi
- Start date
Elon is using the hackathon to find new talent for the FSD team. If Tesla finds some smart new talent that can contribute to the development of FSD, that will be a good thing.
jimmym1981
Member
I think it’s a great idea and I’m one of the biggest FSD believers. We had it reserved for our Y at $6k but for now, it just made sense to spend the extra $2k and get the Performance. I really hope he comes through this year.
diplomat33
Average guy who loves autonomous vehicles
Personally, I think we might see some good fruit from the rewrite this year. But I think next year will be a big year for Tesla's FSD.
By the way, Elon did some big marketing for FSD today:
jimmym1981
Member
That is awesome and I think he finally knows he has it down. I think breakthrough was met and now they can move forward.
Not quite. He was trying to tell us it would never be there, whilst we all acknowledged it wasn't currently there.
I'm encouraged by the fact that Tesla is willing to scrap the old code and start over. I'm also more than ever convinced that my current car will never have the hardware needed for FSD. And there's no doubt at all that I'll have an autonomous car years sooner by buying a new car when FSD is actually in the car than by paying for FSD now and waiting for them to upgrade mine if I'm wrong about its upgradability.
Meanwhile, EAP is very far from perfect, but it's good enough to make driving easier, safer, and more relaxing than if I didn't have it, always assuming that I remain alert, which I do. And with the rewrite going on, it looks as though it will be at least a year before FSD has anything that EAP doesn't other than fancier visualizations on the screen, which I don't care about because I only look at the screen for the speedometer and (if I'm in an unfamiliar area) the map.
Meanwhile, EAP is very far from perfect, but it's good enough to make driving easier, safer, and more relaxing than if I didn't have it, always assuming that I remain alert, which I do. And with the rewrite going on, it looks as though it will be at least a year before FSD has anything that EAP doesn't other than fancier visualizations on the screen, which I don't care about because I only look at the screen for the speedometer and (if I'm in an unfamiliar area) the map.
Cirrus MS100D
Member
I am very pro-FSD (pre-purchased with our 2017 MS) and a strong proponent of Tesla's tech, vision, and capabilities. However, I believe this hack-a-thon (which I think we can all agree is just a marketing stunt for super talent acquisition) is actually a sign that Tesla does not (yet!) have a path towards FSD, possibly acknowledging that the current path is not going to work.
As has been rehashed so many times already, Elon's predictions about when we (customers) would see a tremendous improvement in self-driving has been "off" to put it mildly. If they had a clear path to what they thought would lead to FSD and simply needed more time, more person-hours, or more data, I don't think they would need to be hosting a hack-a-thon for talent scouting. While those items are likely necessary as part of the overall FSD puzzle, I think they are realizing that they may need to rethink some fundamental aspects of their initial plan, and may be hoping to score some new ideas through new talent.
Certainly a lot has changed since they first embarked on this mission (heck, it's been nearly 6 years since AP1 first came out - Kids in high school have now graduated college with math/science degrees!) so it's bound to be good to acquire superstar talent, just out of pure turnover, but I also think this shows they're aware of their own limitations and "need help" pushing further. What I don't quite understand is what NN rockstar that exceeds the high bar isn't already aware of Tesla and what they're doing? Either they've applied and got rejected or have no interest in the job, right?
As has been rehashed so many times already, Elon's predictions about when we (customers) would see a tremendous improvement in self-driving has been "off" to put it mildly. If they had a clear path to what they thought would lead to FSD and simply needed more time, more person-hours, or more data, I don't think they would need to be hosting a hack-a-thon for talent scouting. While those items are likely necessary as part of the overall FSD puzzle, I think they are realizing that they may need to rethink some fundamental aspects of their initial plan, and may be hoping to score some new ideas through new talent.
Certainly a lot has changed since they first embarked on this mission (heck, it's been nearly 6 years since AP1 first came out - Kids in high school have now graduated college with math/science degrees!) so it's bound to be good to acquire superstar talent, just out of pure turnover, but I also think this shows they're aware of their own limitations and "need help" pushing further. What I don't quite understand is what NN rockstar that exceeds the high bar isn't already aware of Tesla and what they're doing? Either they've applied and got rejected or have no interest in the job, right?
diplomat33
Average guy who loves autonomous vehicles
My expectation right now is that the new rewrite with AP3 will probably get to FSD with driver supervision but true driverless FSD will only happen on future hardware. I do think Dojo probably needs 1 year before it will start showing significant results.
We should not look at rewrites as a bad thing. Mostly when working on huge software projects, rewriting large portions of code is a part of the job. This is a good thing, it would be very worrisome if they know the code can be much better but refuse to fix it because "sunk work etc...". Common beginner mistake: It took me hours to write that code, I refuse to delete it, even though I know it could be better.
When I work on large complex software that takes months to complete, I often don't know the ideal structure at the time of beginning. But instead of spending forever trying to get the first attempt perfect, I just start to implement it beginning with what I know how to do. And build step for step from there. I view all that code as pathfinder code.
At some point the full picture will be clear, possibly after testing it in production with real users. At that point I start refactoring and rewriting the portions I don't like, until it's pretty much perfect, and a good starting point for future iterations.
When developing something as large and complex as FSD, it's very likely the finished code will look VERY different than the initial attempts after multiple rewrites and iterations. But each takes you closer, and is a neccessary part of the job.
When I work on large complex software that takes months to complete, I often don't know the ideal structure at the time of beginning. But instead of spending forever trying to get the first attempt perfect, I just start to implement it beginning with what I know how to do. And build step for step from there. I view all that code as pathfinder code.
At some point the full picture will be clear, possibly after testing it in production with real users. At that point I start refactoring and rewriting the portions I don't like, until it's pretty much perfect, and a good starting point for future iterations.
When developing something as large and complex as FSD, it's very likely the finished code will look VERY different than the initial attempts after multiple rewrites and iterations. But each takes you closer, and is a neccessary part of the job.
jimmym1981
Member
To me, I feel Elon needs to re-name this whole FSD because he is working on two different things in my eyes. I think we all know TRUE FSD isn't going to happen on this hardware we have now. The driver-supervision FSD, hopefully, will be out this year. But even then, I have a feeling it will be more user involved than car involved for a good while.
If you look at it like this, how much different is FSD and AP right now? Yes, it takes an exit, but that is no different than AP staying in the lines plus a few extra things that are needed. It uses the sensors to see if a car is there and gets over in a lane. No real different technology than most blind side detection mirrors.
The big hurdle is all of the unknowns in a city that FSD has to deal with, especially a big one like Chicago. People running across the street, double buses, odd-shaped intersections, yellow lights when cars are backed up in intersection, mislabeled street signs, snow on the road, bike lane's that merge...etc. I could go on for days. This is what will take much more driver interaction than anything.
If you look at it like this, how much different is FSD and AP right now? Yes, it takes an exit, but that is no different than AP staying in the lines plus a few extra things that are needed. It uses the sensors to see if a car is there and gets over in a lane. No real different technology than most blind side detection mirrors.
The big hurdle is all of the unknowns in a city that FSD has to deal with, especially a big one like Chicago. People running across the street, double buses, odd-shaped intersections, yellow lights when cars are backed up in intersection, mislabeled street signs, snow on the road, bike lane's that merge...etc. I could go on for days. This is what will take much more driver interaction than anything.
diplomat33
Average guy who loves autonomous vehicles
It would be worse if Tesla knew that they had no viable path to FSD and did nothing about it. A positive way of looking at the hackathon is that Tesla is trying to find talent to make their solution better. Good computer companies will challenge the smartest and brightest to find problems in their product in order to make things even better. Elon wants to find the brightest people who can bring a fresh perspective and identify problems with Tesla's FSD that maybe the Tesla team can't see and help them make it better. It's like having a friend proofread your book. You are so close to the material that it is hard to be objective but a friend who has never seen the manuscript can come in and find plot holes, problems and ways to improve the manuscript that you could not see.
JCB_Arizona
Member
I’m actually a little less confident with the recent developments. If FSD was feature complete (or soon to be), why the sudden and urgent rush to hire people.
diplomat33
Average guy who loves autonomous vehicles
Now might be a good time to remember that the infamous 2016 FSD demo video posted on their website (and still posted last time I checked) proclaimed that the driver was only there "for legal reasons" and that all that remained to be done before FSD would be released was "validation and testing"... and now in 2020 they are throwing out all their code and rewriting it. So by what twisted logic can any of the claims made in 2016 be considered honest?
I am quite confident that many examples of adequately twisted logic will now be offered by the fanboys...
I am quite confident that many examples of adequately twisted logic will now be offered by the fanboys...
diplomat33
Average guy who loves autonomous vehicles
Karpathy responded to a tweet about how the NN is able to see a stop sign in Tesla's AP raw feed video with some interesting tidbits about how the NN works:
"The raw image data is much higher resolution than this mp4. But on top of that the neural net can in some cases be superhuman (e.g. if you "sweep backwards" in time your labels), often due to subtle contextual patterns."
Andrej Karpathy on Twitter
"The raw image data is much higher resolution than this mp4. But on top of that the neural net can in some cases be superhuman (e.g. if you "sweep backwards" in time your labels), often due to subtle contextual patterns."
Andrej Karpathy on Twitter
Daniel in SD
(supervised)
That's not superhuman. Humans see subtle contextual patterns quite well.
The driver is required for legal reasons and you've got to do validation and testing before release. They just didn't mention that they also had to write the software.
Similar threads
