yeah fair call, all manufacturers have been asleep at the wheel. however they are now closing in on tesla. audi, vw, porsche, have a massive ev rollout plan. 70 ev models across the vw group by 2028 from what i read, delivered in 3 phases. so lets call it 23 different cars every 2 and a bit years. i applaud tesla for paving the way, but who can afford a model s? i earn good money and these are still out of reach. tesla made approx 245k cars in 2018 compared with 11M for vw group, which is where vw has the advantage, volume in keeping costs down, and therefore the rrp.
never in my life have I seen a tech company of any sort give a free class on something they created themselves, instead of showing us marketing PR, selling it to us for 10 times the price...
This. Exactly this is why Elon Musk is one of my heros. And frankly talking about implications of his technology, including limitations in such a real an honest way.
This is so they can build market share and have quite literally a datacenter the size of the earth this is insane what they are doing and honestly i can't wait for the future they can profit ridiculously off of these cars imagine selling advertisements based on where you drive and everybody you drove past
They are journalists who are forced to be there in return for a paycheck. The more questions they ask, the better they seem to be at their job. As for the quality of their questions... their journalists. Neither they or their superiors have a full grasp on the subjects.
1:59:19 Teaching the machine with labels. 2:01:39 Teaching the machine with variety of data. 2:04:24 the world is strange. Using real data is better then simulation. 2:05:18 The three most important things for teaching neural networks. 2:06:07 Teaching object detection. 2:08:31 Neural net process. "Data Engine" 2:14:39 Path prediction around corners. "Paths it can't even see" 2:19:19 3D reconstruction using only video. 2:20:45 Neural net vision just as good as radar.
my biggest take away is choosing the right type of data (e.g. the right drivers, diverse scenarios) and fine-tuning based on feedbacks (human intervention and confidence) out weights sheer training data size.
@@bann13 Yes you are right you need the right kind of data. But you also need a lot of it. That's just how Machine Learning works. Thats why Tesla has a leg up on everyone else. They have more data and more edge case data.
So many questions still. Does the fleet have a unified driving persona or are there local flavours? I wonder because there are different local driving behaviours across the world eg in India or China it might be more chaotic than in Germany or Switzerland, different things are socially acceptable. Also does the fleet know all traffic rules across the world or does it just keep accident avoidance "in mind"? Will the network pick up on bad driving habits, like overtaking on the right, which then again is totally fine to do in the UK or Australia?
Thanks. This is what I was looking for. By the way, I'm missing some context here - who is the audience at this talk? Seems a lot more technical than the usual motoring journalists.
That's why I really like Tesla. Other companies have those charismatic speakers who try to get to you with things that they were told, while tesla has those socially awkward people who actually work on the project they are talking about.
And for my money, it's way more interesting to listen to people who know what they are talking about. This is a long boring video of which I watched the entire thing. Because it's fascinating. And it's not a marketing pitch.
not really... they talked about very tesla specifuc stuff. for the actual network you have to know how arithmetics work and how weights work its way more complicated
I was thinking the same thing, but then two years from now, we'll be looking at how much more cool they'll be in yet another two years, etc. etc. etc. I'm gonna support Tesla as much as I can by buying them as often as the wife will let me get away with. :)
These guys are so awkward I love it. They are real and not worried about looking and sounding like they know what they're talking about. They simply do.
@Samurai Shampoo I'm all about not relying on anything and being autonomous; this is why I'm still on my bike, not wanting to rely on an outside source of energy. Don't worry, I'll never rely on an outside brain to drive. But it's a hell of an assistant, plus you don't have to burn fuel to move your ass around, which is brillant.
Can we just all appreciate the fact that they wait a sec, maybe 1.5 seconds before going onto the next subject so sharing it on a specific timestamp / specific subject doesn't make it awkward with some word of the last part cutting off?
1:49:33 "If the simulation fully captured the real world, then it would be a proof that we are living in a simulation" - Elon on why real world data is more important than simulated data. Nailed it
"We have quite a good simulation, but it doesn’t fully capture the real world. If it did, it would be proof that WE were living in a simulation." Classic one from Elon!
Andres explanation was incredible. Like I’m literally shocked by how well he explained such complicated tasks. Wow. No wonder his course at Stanford ending up doing so well lol
I would love to see a Tesla trying to slowly creep through crowds of people and/or livestock. Not being sarcastic, it'd be honestly facinating to see it happen. Coz you need a certain amount of aggression to "project" an opening ahead of you (you want ppl to be scared of being run over). Theres a lot of psychology involved, so perhaps its more a job for their friends at OpenAI.
India doesn't have the road infrastructure everywhere to support this. Musk stated several times that this system works on American roads. This means that road rules have to be implemented. India needs to invest in the basics of infrastructure in order to take advantage of this technology. Failure to do that much means the country cannot keep up with first world countries. If you think I am wrong, you are missing the big picture. We advance when we put down the infrastructure that moves humanity, and proper (we know what constitutes as proper) roads move society forward because it facilitates trade. India needs to stop wasting its time with space exploration and build it's infrastructure now before the raw materials to build it goes up.
@@iLoveTurtlesHaha Thats correct. And while most of Europe has the infrastructure, i am sure we will have bureaucracy problems putting us 5 years behind !
Agreed, the sheer amount of data that their fleet is gathering is staggering. The entire system keeps getting smarter every day! That means that the company with the most data will also get the best insurance rates, and be able to offer the lowest cost. The scary part about AI is that companies like Tesla, Amazon, and Google that are pushing the boundaries will keep getting exponentially smarter and more capable, making it virtually impossible for others to compete. How do you catch up to a network with 1 or 2 million cars? AI is more than a utility like electricity or factories in the industrial revolution - this shift is massive!
I realised after listening to this and some of the questions at the end that some analysts have not got a clue. It is really worrying to think that people will take their advice to make investment decisions. Some of them did not seem to grasp what a step change Teslas have made. Droning on about lidar, gps and high precision maps just proves they did not listen or did not understand.
Holograms, cars that drive themselves, iphones and not to mention self landing space rockets. Yeah 100 years ago you would be burned at the stake for even mentioning any of these. Modern future is already here.
Elon knows Andrej is the single man responsible for developing a solution and infrastructure for FSD. Imagine going to bed every night realizing you will save thousands and then millions of lives going forward. He’s a genius. I hope he gets more and more credit when they turn on the fleet. Team Andrej.
What I love about the Karpathy's presentation is how it reminds us how much we take our own learning for granted. We all had to painstakingly train our minds/learn everything we "know" today... Like the Tesla neural nets, we spent the first few years of our lives knowing pretty much nothing. We had to learn what a dog was and what it looked like, we had to learn how to balance ourselves when we figured out how to stand up... and when we started driving we had to learn how to gauge distances at speed as well as the velocity of other vehicles, and we had to learn how to anticipate where the road would lead when it curved around sharp corners.... (especially at night!) Now we're "just" using brute force to teach computers the same things we've learned ourselves. There are no shortcuts if we want to teach the system to see the world the way we do, and behave appropriately.
I watched this following an article that had it linked and mentioned briefly how Tesla goes all-in with vision only, and the article writer had doubts. Glad I watched it, because he explained it so well, especially the part about unexpected objects, events on roads that no simulation can potentially present; meanwhile Tesla cars actually learn from ALL situations, no matter how rare they are. I have full confidence in vision-based AI now, excellent presentation!
As a computer science major and IT professional I found it fascinating. I didn't major in AI though, so I'll need to read up on some of those neural net terms like ReLU.
@Federal Bureau of Investigation Yes, and my comment was stupid anyway, since even in absolute darkness you can always use a lamp that emits a light that can go through fog or even walls, then detect the reflection with a suited camera, the difference becomes the 3d model extrapolation tech, and i think visual ai will always be cheaper than. Since it measures the distance by computation, like humans, and not by a physical depth sensing mechanism.
So very true, is it bad, i watched all of it, and some parts twice, i imagine to wake up every morning and be involved in projects like this is amazing, no wonder elon sleeps on the couch at tesla :) i would too
I own a Tesla and my mind is blown after watching this video. I use autopilot every single day, but I had no idea HOW it works. Now I know that my car is plugged into the Tesla Matrix and is constantly improving itself. Amazing.
I loved the way Elon handled the introduction of Andrey Karpathy. Just saying he's a Stanford PHD does indeed not make him justice. Andrey is one of the greatest minds involved in AI research right now. I followed (on UA-cam) the classes he held at Stanford and his blog many years ago, before he went to Tesla, and learned almost all the foundation I have about ML/AI from him. He's really passionate about his work, and I'm convinced great things will come from him and his team.
@Samurai Shampoo As Elon said "there's nothing wrong with buying a horse if you want a horse". Lol Of course, he meant that if you want a good electric car, buying the other brands of electric cars is like buying a horse instead of a car.
Horses have pretty good auto-homing ability with predator avoidance, free automatic self-fueling and most cannot even be manually driven into unsafe situations, and they are self-replicating. So in some aspects I think horses will still have the edge.
@@samuvisser it just not necessary to have lidar. With vision methode only and if the neural network mastering it there is no problem. And elon reason is right. It just useless to have lidar at tthose point and waste the money. It's like chopping vegetable with chainsaw
@@samuvisser he pointed out that lidar gives you points of objects in space, car is just a box for it, lidar does not give you information if it moves in reverse or forward, does not read signs or car indicators, so I guess what they trying to say is - trained cameras can give you same/more information that lidar can while are cheaper. This is as I understood it at least. The only thing I could think lidar can be better is and night , outside of the city where camera can be as blind as human eye can be and lidar could see deer crossing the road or some drunk person in black cloths...
1:49:33 "If the simulation fully captured the real world, then it would be a proof that we are living in a simulation" - Elon on why real world data is more important than simulated data. Nailed it
i love my tesla, i love what this guy is doing for humanity, im sold on his company, he is truly and honestly in our timeline a walking genius from another galaxy.
I also really enjoy how he puts down higher education. It brings forth the idea that higher education does not differentiate you from others nearly as much as your experience.
Yeah that guy got it a couple of times, he tried moving on from questions and got shut down then too. Cool how Elon isn't too fussed about qualifications but all jobs on their website require a degree
I recently started working on object detection stuff for the first time ever. This presentation answers many of the real world problems I had like "bicycle attached to the car" kinda situations. This is the best presentation I have ever watched about anything.
I love how technical this is!! I’m studying data science and really appreciate how they share every detail about both the hardware and software. Really interesting!
I really want to get into machine learning and this presentation was very interesting and detailed. I so want to get into the Machine Learning and AI field.
thinking of it, it was a nice event - eventho i dont have a drivers licence i realize the importents of autonomous driving, the application in autonomous transportation and the need for transport systems running on renewable energy rather then my grandma. a shout out to every single employee at tesla and the great minds behind it. Love you elon
This must be their future $25k car. If it has no steering wheel, so there’s no need for performance which every Tesla has. They’ll cut away unnecessary components and specs, and cut hence costs. These are just the obvious conclusions, I’m sure there’ll be more specific features to the car Tesla decides on.
Hopefully not. Driving is incredibly fun, especially performance cars. Level 4 autonomy should be maximum, wheel is there at all times but car can drive you wherever you want.
I love how transparent they are with their tech. This isn’t a marketing presentation, this is a full on course on Tesla’s hardware capabilities. Really mind blowing. Not many companies are pushing us towards the future like Tesla. Tesla is defining the EV market and promoting healthy competition.
What competition!? No one can compete with what I just heard in this presentation. There are no other cars- electric or otherwise- that are anywhere near Tesla's level. Other EV companies may be able to compete with each other, but not with Tesla!
Elon, essentially put all his big guns on stage and let them do what they do best, communicate their expertise on the subject that they know in and out. Did you all notice how simple their slide decks were? It's like... they're just letting the tech they developed speak for itself. Mad boss level.
In the “blue drivable” space shown on the neural net camera presentation, does the system identify potential drivable space in case of an emergency evasive action? Such as, driving into the side of the road if the road is blocked.
regarding snow driving.... (2:40:20) In really heavy snow people tend to only drive where they are familiar (short trips to the shop or to work etc.) so the sample data from the fleet will be biased to snowy drives where the drivers are relying on their memory of the road layout than anything visual. However, the driveable space point is very valid. In extreme snow even without memory the car should be able to function cautiously by seeing snow mounds at the sides of the curbs indicating the driveable space etc.
Its all about how you sell it. Tesla: When we remove your steering wheel, we will replace it with a Margarita machine so you can drink and drive. Crowd: Wait what?!? Take all my money!
That guy at around 3:25:00 asking about HD maps, "we didn't hear much about HD maps..." I guess earlier when Elon was categorically stating that HD maps are not the way to go, this guy was playing with his phone. Why even go to the event if you're gonna waste everyone's time?
Well, there was SO MUCH information throughout the whole presentation... I watched it live (on UA-cam) and I have to admit: I also missed the HD maps part before he asked. Ok, I'm not a native speaker, so it is sometimes difficult to keep up (those guys talked FAST and about very complex topics)... Don't be too harsh to him :)
In his defence, he didnt say that NOTHING was said about HD maps. Perhaps what _was_ said just wasnt enough & he wanted more - he might work in the field himself.
Have to say, Elon was remarkably polite and respectful in his response to the stupid questions on this occasion too. This hasn't always been the case on conference calls in the past...
I am just listening to the discussion of driving in snowy conditions at about 2:40:00. I am reminded of a unique driving event that I encountered a number of years ago. I was driving on a four lane divided highway at about 2 am. There was very little traffic (thankfully) and it was very dark. Hail started to fall. The hailstones bounced off the road surface in such a way that it was impossible to see any aspect of the road. I slowed to around 10 kph. Even at that low speed I basically meandered from one side of the road to the other. I could only sense my wheels going off the road and into the ditch. It did not take me too long to realize that my only sensible option was to park on the shoulder and wait for the hail storm to stop. Hopefully Tesla cars will also have the intelligence to realize when it is simply impossible to drive safely and park themselves in a suitable location and wait. BTW, I am really enjoying this video. It is fascinating.
@Samurai Shampoo You miss the entire point of my post. Right now the software is substantially less capable than a human. Perhaps in future it will equal or best the capability of humans. However this is irrelevant to my point. There will occasionally exist circumstances when it is simply unsafe to drive. The software needs to be aware of when it is not safe to drive and behave accordingly. I live in northern Canada. I have been driving for about 45 years. Many accidents occur because people overestimate their capability in extremely challenging conditions. I would speculate that the conditions that I described (not having any visual access to any part of the surface of the road or shoulder) probably occur very rarely. In 45 years I experienced such conditions only once for a duration of about three minutes. If the scarcity of such events is indeed true, then the neural net will not have any training on such conditions. My impression from the video is that the neural net needs relevant training. No such training is likely to exist. If indeed such conditions were common, then the training it would receive from any reasonable human driver would be to stop and wait until conditions improve.
never going to happen until you end "anomalies" like j walking or pets running. everything has to be linked in and precision sensors without processing delay jesus it will be the same nightmare as people driving. computer reboot would have to happen in miliseconds. not there yet by a mile and a half.
@@paulkim1011 3 camera frames is all that's needed for spatial depth perception. From 60 fps cameras, that's 48 milliseconds; well below even the best human's reaction time.
Buy Tesla stock today.. see it multiply by 2 at the end of 2020. Update: Tesla stocks not only doubled, but tripled and its Aug 2020, not even December yet.
Good Riddance though. Focus on other important problems like fire, health, and security with safer transportation to and from them is a huge improvement. Funding for Police is bad as is, and should be taken care of in other means anyways, right?
Q1. 학습은 어떻게 시킬 것인가? 차량을 주차하고 있으면 되는 것인가?(2:25:54) A1. 비디오 기반, unsupervised 학습기인 Dojo를 이용할 것이다.(2:26:45) Q2. 막대한 양의 데이터 연산을 위해 어떤 것을 고려하고 있는가?(2:37:53) A2. sort of a AWS 이용하는 것도 고려 중이다. (2:38:30)
It's so silly that on Apple events a new camera mode gets whole room clapping and Elon which literally is changing the future for humanity - room is quite
zukacs. That's because it's a tech presentation to engineers, computer scientists and tech journalists. Some, not all, have some understanding of the topics discussed. They are there to report on the progress that Tesla is making in FSD. It is very different to a product launch full of fanboi's and UA-camrs
But the sensor suite DOES need to be changed. Left & Right cameras are preposterously almost 9 FEET behind the nose. Obstructions of foliage/box vans, etc. necessitate the car to creep out fully into danger to see anything. They're also incredibly low quality for something like UPLs that require seeing clearly FAR down the road, etc
The 3D depth map they created from just 6 seconds of video is absolutely incredible to me. I was always skeptical about proceeding into full self-driving without Lidar, but color me convinced.
look up, photogrammetry. you can do it on your computer with a bunch of photos to make a 3d model, but say 100 photos might take you 6 hours of processing to do it. the ability to do in real time would be amazing. they are just showing that pictures can make 3d models, therefore radar is not required if you can see it.
The whole “you need Lidar for self driving” seemed more like something the lidar using company’s wished, not reality. Musks claims always made sense “better sensors will make lidar redundant”, which obviously they are.
@@Scholzey since they can create depth maps via ml in real time now and also know the placement of the cameras on the car, they could probably generate these in minutes
I am a teenager . Studying engineering. I was born and raised with technology. This shouldn't surprise me and should be a new "gadget" . But I'm getting goosebumps because I know how hard it is to do what Tesla achieved . They are bringing sci fi to real life .
1:20:06 I would love to hear this professor's result of applying his architecture design of input output and redundancy to city design (for Mars or the Boring company). He seems to have it coved.....
Brian Smith true, but still. Self Driving is about data, and lots of it. Tesla gets it from the cars they sell, Waymo gets it from their geofenced self driving taxis. I can respect both approaches
Geofencing is absolutely going to be a thing. Rolling this out nationwide is impossible from a legislative point of view. Also, it’s incredibly stupid to not geofence any new AI self driving solution.
@@onemanshow4116 That's true, but what he's saying, is that the car itself will be capable to drive everywhere, but it isn't allowed. It will be capable because currently the cars are already driving there and collecting data. By the time the rules in a specific location allow the car to drive at their locations, it must be capable to do so. The Waymo taxies for example only get data from the part where they are allowed to drive. If you move that car from the US to the UK for example, it won't know how to drive there because it isn't used to driving on the other side of the road (stupid Brattains...). Tesla's however are currently driving in both regions and are collecting data in both regions, so they are capable to drive autonomous in both regions.
@@samuvisser It is the same approach for the learning part but Waymo is learning from way fewer cars in way fewer situations. If they had as much data as Tesla then they would need years to catch up after that point.
History is full of examples where the best technology doesn't always win. Only time will tell but my probability is on Tesla. If only the Simpsons could make an episode to alleviate all doubt. Good news is both determinations are only a function of time.
@@gantmj He said something good about Tesla. Nothing about himself. He is right that this news should have been the biggest thing in years and yet people seem to be sleeping on it.
@@0ooTheMAXXoo0 He did say something about himself. He sees himself as part of the top .1% of perceptive individuals, which in my opinion, is an inflated sense of self in this context.
2:59:45 Elon: "all those lane charges have occurred with zero accidents" Guy: "that is correct, I have watched ever single accident" Elon looks back at him like "what?"
I think the software guy meant that accidents do happen (it's probably impossible to have 0% accident rate), but having watched most of the accident footage, the guy saw that none of these accidents happened during lane changes (could have happened at a traffic light or four-way intersection for example). I still think the guy made a big mistake by saying that though; People should never talk with ambiguity in career defining presentations like this. Humans make mistakes of course but we need to try our best to avoid & learn from such mistakes.
Two years after first watching this, my mind still blows up with all the amazing tech and future vision discussed on this video. It's also wonderful to see the results we're now seeing in FSD 10.8 beta with more to come. This is truly fascinating and inspiring.
I watched the entire thing for first time in my professional life a conference 3+ hours long that is and this has been more exciting than a blockbuster movie. Elon you the man! Hats off to you Sir and to your vision and tenacity! Tenacity, tenacity, tenacity!
3:03:17
"There's still in 2019 no car that can compete with Model S of 2012, it's 7 years later... still waiting"
I love it
That eyeroll was majestic
Collin The Viking ikr!
"Reach S3X production of 10000/week"
yeah fair call, all manufacturers have been asleep at the wheel. however they are now closing in on tesla. audi, vw, porsche, have a massive ev rollout plan. 70 ev models across the vw group by 2028 from what i read, delivered in 3 phases. so lets call it 23 different cars every 2 and a bit years. i applaud tesla for paving the way, but who can afford a model s? i earn good money and these are still out of reach. tesla made approx 245k cars in 2018 compared with 11M for vw group, which is where vw has the advantage, volume in keeping costs down, and therefore the rrp.
ozzycarnut with time, Teslas will cost $25K or less like Elon said
Also, by 2025 Tesla will probably be the 1st company to create a flying car lol
never in my life have I seen a tech company of any sort give a free class on something they created themselves, instead of showing us marketing PR, selling it to us for 10 times the price...
This. Exactly this is why Elon Musk is one of my heros. And frankly talking about implications of his technology, including limitations in such a real an honest way.
@@gwentarinokripperinolkjdsf683 49 people agree with me, none with you.. yeah, lmao..
Yup, Tesla isn't into marketing PR at all. Got your solar roof yet? Level 5 autonomy?
They’re trying to convince politicians
This is so they can build market share and have quite literally a datacenter the size of the earth this is insane what they are doing and honestly i can't wait for the future they can profit ridiculously off of these cars imagine selling advertisements based on where you drive and everybody you drove past
1:32:27 - Chief engineer cites performance increase of 2100% - not a single clap in the room...
Yeah, no one has even a remote personality in that room, apart from Elon and the engineers.
They are journalists who are forced to be there in return for a paycheck. The more questions they ask, the better they seem to be at their job. As for the quality of their questions... their journalists. Neither they or their superiors have a full grasp on the subjects.
@@durimmiziraj4815 I think... they're investors? Not journalists.
@@rope8666 My mistake, your right. Well that explains the nonchalance then.
Both journalists and investors
Greetings 1:09:33
1 of 3 System Arch - Peter Bannon 1:12:08
- Getting Started 1:12:38
- Full self-driving computer 1:14:13
- Full self-driving chip 1:19:44
- Neural network accelerator 1:24:07
- Results 1:31:30
-- Conclusion 1:33:02
- Q&A 1:35:11
2 of 3 Neural Net Vision - André Karpati 1:51:05
- Introduction 1:52:05
- Primer in Neural Network 1:54:20
- Fleet Training 2:05:26
- Exit from Lidar 2:16:46
- Q&A 2:25:20
3 of 3 Software Engineering at scale w/Stewart 2:50:09
- Autopilot 2:52:37
- Master Plan - Elon 3:02:08
-- Robo-Taxi 3:05:19
Final Q&A 3:14:33
End 3:44:17
Thanks!!
thx
Thank youuuuu
Thanksss
1:59:19 Teaching the machine with labels.
2:01:39 Teaching the machine with variety of data.
2:04:24 the world is strange. Using real data is better then simulation.
2:05:18 The three most important things for teaching neural networks.
2:06:07 Teaching object detection.
2:08:31 Neural net process. "Data Engine"
2:14:39 Path prediction around corners. "Paths it can't even see"
2:19:19 3D reconstruction using only video.
2:20:45 Neural net vision just as good as radar.
for data scientists or ML enthusiasts, this is the most useful comment!
my biggest take away is choosing the right type of data (e.g. the right drivers, diverse scenarios) and fine-tuning based on feedbacks (human intervention and confidence) out weights sheer training data size.
@@bann13 Yes you are right you need the right kind of data. But you also need a lot of it. That's just how Machine Learning works. Thats why Tesla has a leg up on everyone else. They have more data and more edge case data.
Bump
So many questions still. Does the fleet have a unified driving persona or are there local flavours? I wonder because there are different local driving behaviours across the world eg in India or China it might be more chaotic than in Germany or Switzerland, different things are socially acceptable. Also does the fleet know all traffic rules across the world or does it just keep accident avoidance "in mind"? Will the network pick up on bad driving habits, like overtaking on the right, which then again is totally fine to do in the UK or Australia?
Video starts at
1:09:10
1:52:05 second segment.
Seen times several times just thought it was at something cool but not the start!
Thanks. This is what I was looking for. By the way, I'm missing some context here - who is the audience at this talk? Seems a lot more technical than the usual motoring journalists.
@@utuber2 I think it's a shareholder/stakeholder meeting or something like this.
Conference
you're doing God's work young ser
Thank you for this comment
elon makes the most impressive statements
audience doesn't even clap
elon goes like "I'll say that again.... "
audience:
its one thing saying we think it will be able to do this and then actually doing it.
Because they're investors and it's not the first time they've had his bs.
It's just that we have gotten so used to Elon making the most impressive statements...
Bruno Mailly like sending 2 astronauts to space?
@@musaran2 but here it's literally not "we will try to do that in the future" but "we did that already until today".
Start/Autopilot Hardware w/Stewart - 1:09:11
Neural Net Vision w/André - 1:51:05
Software Engineering at scale w/Stewart - 2:50:09
That's why I really like Tesla.
Other companies have those charismatic speakers who try to get to you with things that they were told, while tesla has those socially awkward people who actually work on the project they are talking about.
Love this comment. Explains the difference perfectly
And for my money, it's way more interesting to listen to people who know what they are talking about. This is a long boring video of which I watched the entire thing. Because it's fascinating. And it's not a marketing pitch.
@Politically correct speech isn't my thing, fk off Damn, are you triggered son?
This is like when you're in an graduate school lecture. Others are like you're in a political rally.
@@CalvinHikes It's a boring video that is fascinating?
Did Tesla just start teaching us how neural networks work? Thanks for the free class. I’m not complaining!
not really... they talked about very tesla specifuc stuff. for the actual network you have to know how arithmetics work and how weights work its way more complicated
Elon: the matrix if you will
Andrej: this is my talk, shut up
@@cinialvespow1054 this book is free to read online and highly regarded as the "bible" of deep learning textbooks: www.deeplearningbook.org/
I think the jeopardy guy has a neural uplink implant to the cloud. Or else just really good
@@ken-mb5cp You don't have the slightest clue what you're talking about lmao. You just threw out as many buzzwords as you could think of.
I refuse to buy a new car for at least two years. Then I will get to own the most epic tesla ever.
I was thinking the same thing, but then two years from now, we'll be looking at how much more cool they'll be in yet another two years, etc. etc. etc. I'm gonna support Tesla as much as I can by buying them as often as the wife will let me get away with. :)
@@floorpizza8074 Every dollar out of your pocket, is ultimately a dollar into a billionaire's pocket. Buy a Tesla, but do it wisely.
imlkrsfn wrong
Elon has to split profit for employee’s salary and he gets a part of it
2 years later, can you give us an update?
Did you buy a tesla?
These guys are so awkward I love it. They are real and not worried about looking and sounding like they know what they're talking about. They simply do.
This is the most significant announcement Tesla has made in its history.
Agreed and it went by virtually unnoticed .
people dont seem to understand that. stocks go down and people rant about the Q1 loss.
Totally agree. This is a new world!
I feel in 20 years this will be a classic important video. Groundbreaking regardless if Tesla survives as a company.
I agree. The moment it was undeniable that Tesla is a fraud
I'm 35 and never owned a car. Living in a big city doesn't require one. But I'm moving out of the city soon.
My first car will be a Tesla.
Tesla is a great investment!
@Samurai Shampoo I'm all about not relying on anything and being autonomous; this is why I'm still on my bike, not wanting to rely on an outside source of energy.
Don't worry, I'll never rely on an outside brain to drive. But it's a hell of an assistant, plus you don't have to burn fuel to move your ass around, which is brillant.
I hope you have a freaking amazing job 😂
@@kissingfrogs106 My job has a link with your profile picture. :)
@@fl260 oh so life sucks for you
Can we just all appreciate the fact that they wait a sec, maybe 1.5 seconds before going onto the next subject so sharing it on a specific timestamp / specific subject doesn't make it awkward with some word of the last part cutting off?
144 trillion operations per second how can I wrap my head around that 🤯
Lol its brok!
wow this was from when he was good.
1:49:33
"If the simulation fully captured the real world, then it would be a proof that we are living in a simulation"
- Elon on why real world data is more important than simulated data.
Nailed it
you need to be a computer with a Tesla chip.
If you can drive, you’re already in excess of whatever TeraFlipFlops they’re using 🥰 🧠 🚘 🖥
"We have quite a good simulation, but it doesn’t fully capture the real world. If it did, it would be proof that WE were living in a simulation."
Classic one from Elon!
Thanks for this epic sentence.
Jithesh M Maybe we are living in a self driving car training simulation.
Literally one sec after reading your comment Elon was saying this exact sentence... That tripped me out haha!
I don't understand how Tesla stock isn't at an all-time high.
Flat-Earthers, climate change deniers, Holocaust deniers and Tesla shorts work together to ruin a good world...
@@2nd3rd1st agree
Because "average" human intelligence is not able to understand this lecture. And "average" people buy stocks.
Buy some while you can then
Unions, oil companies, city officials, taxi drivers, shorts, anti capitalists, and those just jealous.
Andres explanation was incredible. Like I’m literally shocked by how well he explained such complicated tasks. Wow. No wonder his course at Stanford ending up doing so well lol
Guy:
We didn't hear much today around the importance of HD maps... (3:24:53)
Same Guy:
Asleep for 2:38:40-2:43:13
If Tesla can do full autonomy in India, it can do it safely anywhere else in the world.
I would love to see a Tesla trying to slowly creep through crowds of people and/or livestock. Not being sarcastic, it'd be honestly facinating to see it happen. Coz you need a certain amount of aggression to "project" an opening ahead of you (you want ppl to be scared of being run over). Theres a lot of psychology involved, so perhaps its more a job for their friends at OpenAI.
India doesn't have the road infrastructure everywhere to support this. Musk stated several times that this system works on American roads. This means that road rules have to be implemented. India needs to invest in the basics of infrastructure in order to take advantage of this technology. Failure to do that much means the country cannot keep up with first world countries. If you think I am wrong, you are missing the big picture. We advance when we put down the infrastructure that moves humanity, and proper (we know what constitutes as proper) roads move society forward because it facilitates trade. India needs to stop wasting its time with space exploration and build it's infrastructure now before the raw materials to build it goes up.
@@iLoveTurtlesHaha Unfortunately it is not a one or the other, both are essential for a prosperous future
@@iLoveTurtlesHaha Thats correct. And while most of Europe has the infrastructure, i am sure we will have bureaucracy problems putting us 5 years behind !
@@iLoveTurtlesHaha not infrastructure, people lack common sense and take things for granted.
This is perhaps the most significant video presentation in the history of the auto industry. Amazing.
1:29:44 an important remark by Elon
Absolutely. The other auto makers should be shitting their pants
yes... and all the other auto manufacturers are listening very closely . Tesla is 3 steps ahead in E-car tech
Agreed, the sheer amount of data that their fleet is gathering is staggering. The entire system keeps getting smarter every day! That means that the company with the most data will also get the best insurance rates, and be able to offer the lowest cost. The scary part about AI is that companies like Tesla, Amazon, and Google that are pushing the boundaries will keep getting exponentially smarter and more capable, making it virtually impossible for others to compete. How do you catch up to a network with 1 or 2 million cars? AI is more than a utility like electricity or factories in the industrial revolution - this shift is massive!
I realised after listening to this and some of the questions at the end that some analysts have not got a clue. It is really worrying to think that people will take their advice to make investment decisions. Some of them did not seem to grasp what a step change Teslas have made. Droning on about lidar, gps and high precision maps just proves they did not listen or did not understand.
I don’t even own a Tesla and I can’t get enough of these videos. Future is *almost* here!
Holograms, cars that drive themselves, iphones and not to mention self landing space rockets. Yeah 100 years ago you would be burned at the stake for even mentioning any of these. Modern future is already here.
Never seen Elon showing such an amount of respect as he did against Andrej. It seems he's literally adoring him.
well, he hates authority, meaningless titles and paper only "status" like a Stanford PhD
That's a real leader right there. Love Elon!
For a good reason. It's so amazing what they've done!
Elon knows Andrej is the single man responsible for developing a solution and infrastructure for FSD. Imagine going to bed every night realizing you will save thousands and then millions of lives going forward. He’s a genius. I hope he gets more and more credit when they turn on the fleet. Team Andrej.
1:09:10
thanks
Ty
Thank you!
To the top with this!
Jacob Daniel Fernandez On this day you became our hero. The people’s hero. Thanks
What I love about the Karpathy's presentation is how it reminds us how much we take our own learning for granted. We all had to painstakingly train our minds/learn everything we "know" today... Like the Tesla neural nets, we spent the first few years of our lives knowing pretty much nothing. We had to learn what a dog was and what it looked like, we had to learn how to balance ourselves when we figured out how to stand up... and when we started driving we had to learn how to gauge distances at speed as well as the velocity of other vehicles, and we had to learn how to anticipate where the road would lead when it curved around sharp corners.... (especially at night!) Now we're "just" using brute force to teach computers the same things we've learned ourselves. There are no shortcuts if we want to teach the system to see the world the way we do, and behave appropriately.
I watched this following an article that had it linked and mentioned briefly how Tesla goes all-in with vision only, and the article writer had doubts.
Glad I watched it, because he explained it so well, especially the part about unexpected objects, events on roads that no simulation can potentially present; meanwhile Tesla cars actually learn from ALL situations, no matter how rare they are. I have full confidence in vision-based AI now, excellent presentation!
How you know it's an Elon Musk presentation. "An order of magnitude".
He is a business magnate🤣
In the unofficial live chat there was a guy who suggested this as a drinking game. Everytime Elon says "order of magnitude" you need to drink
And some times that is true! Why I love Tesla
Yea thats true but remember that its a common expression in science, its not just him.
It's also the name of a globular album (it's on UA-cam), the album is amazing!
I've never listened to something for this long, with such interest and so little understanding. And yet, I listened to it all.
Andrew Martin me too!
I did as well. Fascinated the whole time.
Andrew Martin nope skipped a huge part of it not a car man
As a computer science major and IT professional I found it fascinating. I didn't major in AI though, so I'll need to read up on some of those neural net terms like ReLU.
Watched it all. Watched it all again.
About LIDAR: If you want to design a bat, use LIDAR. If you want to emulate a human, use computer vision/neural network.
but night?! fog?! I don't get it.
I'm not saying LIDAR is the only solution, but the visible spectrum hides alot at least.
@Federal Bureau of Investigation
Yes, and my comment was stupid anyway, since even in absolute darkness you can always use a lamp that emits a light that can go through fog or even walls, then detect the reflection with a suited camera, the difference becomes the 3d model extrapolation tech, and i think visual ai will always be cheaper than. Since it measures the distance by computation, like humans, and not by a physical depth sensing mechanism.
I love how the roadster comes sweeping by the rest of the cars. Showing us how fast it really is.
I will hesitate to watch a 3 hours movie 🎥 .... but a lecture from Tesla 😀👍
So real :D
haha this is so true. Watched this with intense interest when it was released but can't keep attention to a movie at home today.
So very true, is it bad, i watched all of it, and some parts twice, i imagine to wake up every morning and be involved in projects like this is amazing, no wonder elon sleeps on the couch at tesla :) i would too
Fun fact: Data from Star Trek can perform 60 trillion operations per second which means this car can perform over twice as many operations as Data.
Wonderful fact and shows how much technology has moved on. So in the future Data will basically be the slowest computer in the universe
Mind blown. Thank you for pointing that out.
R u a huge nerd?
@@oalmannearn6949 The hugest nerds happen to also be the richest men on earth.
I own a Tesla and my mind is blown after watching this video. I use autopilot every single day, but I had no idea HOW it works. Now I know that my car is plugged into the Tesla Matrix and is constantly improving itself. Amazing.
Sure~ua-cam.com/video/A1qNdHPyHu4/v-deo.html
If we want to invest in this program how do we start?
Just buy some cars?
Yup. :)
I loved the way Elon handled the introduction of Andrey Karpathy. Just saying he's a Stanford PHD does indeed not make him justice. Andrey is one of the greatest minds involved in AI research right now. I followed (on UA-cam) the classes he held at Stanford and his blog many years ago, before he went to Tesla, and learned almost all the foundation I have about ML/AI from him. He's really passionate about his work, and I'm convinced great things will come from him and his team.
time stamp of when andrey comes on?
@@user-zn3wm7pb3w 1:52:09
Absolutely agree
Bae: I'm home alone, come ove...
*Tesla livestream pops up*
Me: *Hangs up*
Bae: *Comes over to watch livestream*
Bae: *is MUSK*
LOL! yup
Tesla robo sex bae coming 2069
@@DaygoG More like 2030 plus or minus 5 ...but who is counting.
Delivery day August 6th 12pm thanks Tesla of fremont an the team at Cleveland Tesla!!! 😁😁😁 especially Elon making my dream come true 👍
I wonder why 1.5K people dislike something as awesome as this
3:29:35 "If you're not buying a Tesla, it'll be like owning a horse in 3 years." Made my day
As in, "If you don't buy a Tesla, then you will be outdated"?
@Samurai Shampoo Ok.
@Samurai Shampoo As Elon said "there's nothing wrong with buying a horse if you want a horse". Lol
Of course, he meant that if you want a good electric car, buying the other brands of electric cars is like buying a horse instead of a car.
Horses have pretty good auto-homing ability with predator avoidance, free automatic self-fueling and most cannot even be manually driven into unsafe situations, and they are self-replicating. So in some aspects I think horses will still have the edge.
@@cspanza Horses have self fueling? Ever feed a horse $$$ Ill stick with the super charging stations for free. lol
Before Andrej's presentation, I was 90% convinced of Lidar-less autonomy. Now I am 99.9999% convinced.
And that is why 0.0001% of people are right
Just because Elon said so? He gave 0 reasons except for cost. And that can always change with smart engineering. I don't know man, not convinced yet
@@samuvisser it just not necessary to have lidar. With vision methode only and if the neural network mastering it there is no problem. And elon reason is right. It just useless to have lidar at tthose point and waste the money. It's like chopping vegetable with chainsaw
@@samuvisser he pointed out that lidar gives you points of objects in space, car is just a box for it, lidar does not give you information if it moves in reverse or forward, does not read signs or car indicators, so I guess what they trying to say is - trained cameras can give you same/more information that lidar can while are cheaper. This is as I understood it at least. The only thing I could think lidar can be better is and night , outside of the city where camera can be as blind as human eye can be and lidar could see deer crossing the road or some drunk person in black cloths...
The last few 9's are gonna be hard..
Proud owner of a 2021 Model 3 LR AWD the future is moving quickly absolutely love our Tesla.
Elon’s fundamental understanding of processes make him the innovation king of the century
1:49:33
"If the simulation fully captured the real world, then it would be a proof that we are living in a simulation"
- Elon on why real world data is more important than simulated data.
Nailed it
This is probably one of the most important videos for anyone learning and practising Computer Vision .
not mine
Greetings 1:09:33
1 of 3 System Arch - Peter Bannon 1:12:08
- Getting Started 1:12:38
- Full self-driving computer 1:14:13
- Full self-driving chip 1:19:44
- Neural network accelerator 1:24:07
- Results 1:31:30
-- Conclusion 1:33:02
- Q&A 1:35:11
2 of 3 Neural Net Vision - André Karpati 1:51:05
- Introduction 1:52:05
- Primer in Neural Network 1:54:20
- Fleet Training 2:05:26
- Exit from Lidar 2:16:46
- Q&A 2:25:20
3 of 3 Software Engineering at scale w/Stewart 2:50:09
- Autopilot 2:52:37
- Master Plan - Elon 3:02:08
-- Robo-Taxi 3:05:19
Final Q&A 3:14:33
End 3:44:17
i love my tesla, i love what this guy is doing for humanity, im sold on his company, he is truly and honestly in our timeline a walking genius from another galaxy.
The way that Elon nonchalantly spits facts gets me hyped.
The fake facts for the groupies.
He's done it at another analyst day. People forget he has a degree in physics and economics.
This men are gods of science
His quantum mechanics refresher did split photonic wavelengths correctly.
@@blanamaxima GET LOST IDIOT
Liked how Elon said something like "There's a lot of people with PhDs from Stanford, Andrej tell us about the actual cool stuff you have done"
It seems like Elon was getting really annoyed with that guy
cumquat nah, it's just with badges of merit.
I also really enjoy how he puts down higher education. It brings forth the idea that higher education does not differentiate you from others nearly as much as your experience.
Yeah that guy got it a couple of times, he tried moving on from questions and got shut down then too. Cool how Elon isn't too fussed about qualifications but all jobs on their website require a degree
@@keeparisicecream6156 I don't think anybody is denying he has a PhD
I recently started working on object detection stuff for the first time ever. This presentation answers many of the real world problems I had like "bicycle attached to the car" kinda situations. This is the best presentation I have ever watched about anything.
Because I drive a Tesla, I work for Tesla,, so that Tesla will soon work for me.
Tesla taught me the basics of Deep Machine Learning and how AI works from one of the Stanford professors in 2 hours. Thanks Tesla.
totally agreed, best explanation online yet. no question these guys know what they're doing...
Andrej Karpathy is amazing!
YES & EXCELLENT.
Best part is you did not have to take out an $100,000 loan to load this video.
Which course? Can you post a link?
I love how technical this is!! I’m studying data science and really appreciate how they share every detail about both the hardware and software. Really interesting!
Ditto
same here and that dude discussing FAD and neural network was quite good
Yea I'm glad they showed us the network architecture, I've always wondered what they use. And now I know it's a medium sized inception model. :)
Nerds
I really want to get into machine learning and this presentation was very interesting and detailed. I so want to get into the Machine Learning and AI field.
thinking of it, it was a nice event - eventho i dont have a drivers licence i realize the importents of autonomous driving, the application in autonomous transportation and the need for transport systems running on renewable energy rather then my grandma.
a shout out to every single employee at tesla and the great minds behind it.
Love you elon
One yr later..........most valuable car company in the world.....10 yrs later....most valuable company in the world!!!
@Stefan D. initial AI learning requires a lot of manual inputs & correction, its all about resource management to max market penetration
“In 3 years time we will be having cars without steering wheels.”
I’ll be back to this comment in 2022 😅
If we still exist by then
This must be their future $25k car. If it has no steering wheel, so there’s no need for performance which every Tesla has. They’ll cut away unnecessary components and specs, and cut hence costs. These are just the obvious conclusions, I’m sure there’ll be more specific features to the car Tesla decides on.
@john smith pessimists you never cease to amaze me with your stupidity. aand now we wait for the aggressive response.
Take away my steering wheel, and I'll buy a gas powered car.
Hopefully not. Driving is incredibly fun, especially performance cars. Level 4 autonomy should be maximum, wheel is there at all times but car can drive you wherever you want.
I love how transparent they are with their tech. This isn’t a marketing presentation, this is a full on course on Tesla’s hardware capabilities. Really mind blowing.
Not many companies are pushing us towards the future like Tesla. Tesla is defining the EV market and promoting healthy competition.
What competition!? No one can compete with what I just heard in this presentation.
There are no other cars- electric or otherwise- that are anywhere near Tesla's level. Other EV companies may be able to compete with each other, but not with Tesla!
Elon is smart
Elon, essentially put all his big guns on stage and let them do what they do best, communicate their expertise on the subject that they know in and out. Did you all notice how simple their slide decks were? It's like... they're just letting the tech they developed speak for itself. Mad boss level.
Ok you attended this course. Tell me how much compute, networking and storage there is in every Tesla Model 3 car.
In the “blue drivable” space shown on the neural net camera presentation, does the system identify potential drivable space in case of an emergency evasive action? Such as, driving into the side of the road if the road is blocked.
regarding snow driving.... (2:40:20)
In really heavy snow people tend to only drive where they are familiar (short trips to the shop or to work etc.) so the sample data from the fleet will be biased to snowy drives where the drivers are relying on their memory of the road layout than anything visual.
However, the driveable space point is very valid. In extreme snow even without memory the car should be able to function cautiously by seeing snow mounds at the sides of the curbs indicating the driveable space etc.
Elon does loosely touch on this a little at 2:41:20
Apple: we put more pixels in the iphone
Crowd: OHHHMAGASH UR DA BEST!!!!!
Tesla: we will have a car that drives itself by next year
Crowd: meh.
For a lot of sheeple, the iphone takes up 98% of their lives
Its all about how you sell it.
Tesla: When we remove your steering wheel, we will replace it with a Margarita machine so you can drink and drive.
Crowd: Wait what?!? Take all my money!
Apple: we made it easier for thots
Is OHMAGARSH in the way Conan O'brien would say it?
Its because people don't believe it will happen. I hope it does though.
That guy at around 3:25:00 asking about HD maps, "we didn't hear much about HD maps..." I guess earlier when Elon was categorically stating that HD maps are not the way to go, this guy was playing with his phone. Why even go to the event if you're gonna waste everyone's time?
Ha, that was my exact thought at the time of watching too. Would have felt like shouting "Keep up at the back" if I was there haha
Well, there was SO MUCH information throughout the whole presentation... I watched it live (on UA-cam) and I have to admit: I also missed the HD maps part before he asked. Ok, I'm not a native speaker, so it is sometimes difficult to keep up (those guys talked FAST and about very complex topics)... Don't be too harsh to him :)
In his defence, he didnt say that NOTHING was said about HD maps. Perhaps what _was_ said just wasnt enough & he wanted more - he might work in the field himself.
Have to say, Elon was remarkably polite and respectful in his response to the stupid questions on this occasion too. This hasn't always been the case on conference calls in the past...
To the person that watch this all the way to the end without skipping. You are now the most dedicated person ever
I am just listening to the discussion of driving in snowy conditions at about 2:40:00. I am reminded of a unique driving event that I encountered a number of years ago. I was driving on a four lane divided highway at about 2 am. There was very little traffic (thankfully) and it was very dark. Hail started to fall. The hailstones bounced off the road surface in such a way that it was impossible to see any aspect of the road. I slowed to around 10 kph. Even at that low speed I basically meandered from one side of the road to the other. I could only sense my wheels going off the road and into the ditch. It did not take me too long to realize that my only sensible option was to park on the shoulder and wait for the hail storm to stop. Hopefully Tesla cars will also have the intelligence to realize when it is simply impossible to drive safely and park themselves in a suitable location and wait. BTW, I am really enjoying this video. It is fascinating.
@Samurai Shampoo You miss the entire point of my post. Right now the software is substantially less capable than a human. Perhaps in future it will equal or best the capability of humans. However this is irrelevant to my point. There will occasionally exist circumstances when it is simply unsafe to drive. The software needs to be aware of when it is not safe to drive and behave accordingly. I live in northern Canada. I have been driving for about 45 years. Many accidents occur because people overestimate their capability in extremely challenging conditions. I would speculate that the conditions that I described (not having any visual access to any part of the surface of the road or shoulder) probably occur very rarely. In 45 years I experienced such conditions only once for a duration of about three minutes. If the scarcity of such events is indeed true, then the neural net will not have any training on such conditions. My impression from the video is that the neural net needs relevant training. No such training is likely to exist. If indeed such conditions were common, then the training it would receive from any reasonable human driver would be to stop and wait until conditions improve.
Full self driving cars will be an amazing reality. Can't wait to see it in the near future. This was a great tech event in a long while.
The future is already here. Tesla just showed a full self driving computer lol
never going to happen until you end "anomalies" like j walking or pets running. everything has to be linked in and precision sensors without processing delay jesus it will be the same nightmare as people driving. computer reboot would have to happen in miliseconds. not there yet by a mile and a half.
lover and a hater it is already happening.
lol.
@@paulkim1011 3 camera frames is all that's needed for spatial depth perception. From 60 fps cameras, that's 48 milliseconds; well below even the best human's reaction time.
Proud Tesla Owner right here! Seriously this company is insane!
I hope to one day own a Tesla, and maybe even work for them! What Model do you have?
I am 61. It's great to be living in this time with Elon Musk
And going broke
Gamer Gey. Give me ur Tesla!
B J you must be a short seller
bro i was watching pov car drive but fell asleep and woke up here at 3 hour
Me encanta Tesla, sus diseños , su seguridad, su comodidad y su aceleración y además su tecnología es uff Viva Tesla!!!
Andrej Karpathy is from x1.25 world simulation.
Playing it at .75x is clear. Worth understing what he says! But it just have been hard for the people who were there, live.
Buy Tesla stock today.. see it multiply by 2 at the end of 2020.
Update: Tesla stocks not only doubled, but tripled and its Aug 2020, not even December yet.
By 2 would be laughably poor. I expect at least 3 by end of 2020
you were right too early, it will probably be much more in 2020 Dec
Well, it doubled already.
the oralcy its 460
900
*Lets be honest we came here for the picture in the thumbnail of the video*
Congratulations Tesla for 2 million subscribers
RIP police departments that source their revenue off of speeding tickets
Good Riddance though. Focus on other important problems like fire, health, and security with safer transportation to and from them is a huge improvement. Funding for Police is bad as is, and should be taken care of in other means anyways, right?
@@cudlzzz567 how is not funding the police a good idea!? Sure all those jobs are important but so is the police.
FVViolins 00 like everyone is gonna have a Tesla?? Small chance
Rui Wang wow never knew that... Good point, I need to do more research lol
lol they'll find some way to harass citizens
the speaker at 1:52 and onwards describing neural networks did a good job of explaining things clearly
1:52:00
Yes he's Andrej Karpathy
He used to teach that at Stanford
@@DaveHunterDave yeah lol. I set playback speed to .75x and I was able to take in much more.
Such an amazing presentation from such knowledgeable people. I look forward to the future of Tesla and the amazing leadership from Elon Musk.
Q1. 학습은 어떻게 시킬 것인가? 차량을 주차하고 있으면 되는 것인가?(2:25:54)
A1. 비디오 기반, unsupervised 학습기인 Dojo를 이용할 것이다.(2:26:45)
Q2. 막대한 양의 데이터 연산을 위해 어떤 것을 고려하고 있는가?(2:37:53)
A2. sort of a AWS 이용하는 것도 고려 중이다. (2:38:30)
It's so silly that on Apple events a new camera mode gets whole room clapping and Elon which literally is changing the future for humanity - room is quite
That's because pretty much everyone, except maybe the Sikh guy up front, has no idea what Elon and his team are talking about.
zukacs haha they’re all confused
zukacs. That's because it's a tech presentation to engineers, computer scientists and tech journalists. Some, not all, have some understanding of the topics discussed. They are there to report on the progress that Tesla is making in FSD. It is very different to a product launch full of fanboi's and UA-camrs
@@lsloan33 So true !!! Thats why this is on UA-cam as well ;)
Most of the Apple event viewers are probably paid actors 😆
1:55 Did this car just drove through my windows XP background??
underrated
😂
The Windows XP background is now a dead, barren landscape in real life.
Dead and barren? So, ... a grapeyard, then?
0:29 Track ID, please. Fantastic sound.
But the sensor suite DOES need to be changed. Left & Right cameras are preposterously almost 9 FEET behind the nose. Obstructions of foliage/box vans, etc. necessitate the car to creep out fully into danger to see anything.
They're also incredibly low quality for something like UPLs that require seeing clearly FAR down the road, etc
I reckon cameras on the bumpers pointing to the right and left would be great on all cars. I have idle thoughts about installing it on my car.
@@danielstapler4315 It's like we're separated at birth
The 3D depth map they created from just 6 seconds of video is absolutely incredible to me. I was always skeptical about proceeding into full self-driving without Lidar, but color me convinced.
@2:18:50
i wonder how did you always imagine reading signs with lidar
look up, photogrammetry. you can do it on your computer with a bunch of photos to make a 3d model, but say 100 photos might take you 6 hours of processing to do it. the ability to do in real time would be amazing. they are just showing that pictures can make 3d models, therefore radar is not required if you can see it.
The whole “you need Lidar for self driving” seemed more like something the lidar using company’s wished, not reality. Musks claims always made sense “better sensors will make lidar redundant”, which obviously they are.
@@Scholzey since they can create depth maps via ml in real time now and also know the placement of the cameras on the car, they could probably generate these in minutes
Tesla is so damn cool
Kopymatic, my hearing enhanced after I heard “He was the lead designer for the Apple iPhone 5 just before joining Tesla.
I like when they bust into flames
@@raymeyers8554 So you must love the hundreds of thousands of ICE vehicles when they burst into flames each year... just in the US.
I am a teenager .
Studying engineering.
I was born and raised with technology.
This shouldn't surprise me and should be a new "gadget" .
But I'm getting goosebumps because I know how hard it is to do what Tesla achieved .
They are bringing sci fi to real life .
You may think why is there no sound? Well it’s a god damn TESLA!
I feel stupid and smart after watching this
More stupid than smart
tubetop123 hhhhhhhh
I’ve learned some stuff with my own neural network. :)
That's what real learning feels like.
I’m feeling kind of ...jobless😭. I’m a taxi driver.
This is better than any science fiction story I've ever read and it's reality!
Bible?
@@sadbutitstrue2323 I get the fiction there, but where's the science?
Ben Snow damn
Ben Snow don’t ignore the guy who told you to read more
If you want more science fiction, check the human brain project and the spinnaker computer.
1:20:06 I would love to hear this professor's result of applying his architecture design of input output and redundancy to city design (for Mars or the Boring company). He seems to have it coved.....
Everyone was so quite in this event because no one knew how to react. Elon is doing things which no one grasps the scope of
"Mark my words" loving it when he says this.
....both a help and a hindrance. "Measured" words are quite helpful. wink.
I believe it is not just me, personally I super appreciate the technical detail explanation. Thank you Tesla & Elon
2 years later and this still blows away anything that other OEMs are doing. Everyone else should be watching this and taking notes.
The initial montage of Tesla cars is just amazing!!!
"if you need a geofenced area, you don't have real self driving."
Brian Smith true, but still. Self Driving is about data, and lots of it. Tesla gets it from the cars they sell, Waymo gets it from their geofenced self driving taxis. I can respect both approaches
Geofencing is absolutely going to be a thing. Rolling this out nationwide is impossible from a legislative point of view. Also, it’s incredibly stupid to not geofence any new AI self driving solution.
@@onemanshow4116 That's true, but what he's saying, is that the car itself will be capable to drive everywhere, but it isn't allowed. It will be capable because currently the cars are already driving there and collecting data. By the time the rules in a specific location allow the car to drive at their locations, it must be capable to do so. The Waymo taxies for example only get data from the part where they are allowed to drive. If you move that car from the US to the UK for example, it won't know how to drive there because it isn't used to driving on the other side of the road (stupid Brattains...). Tesla's however are currently driving in both regions and are collecting data in both regions, so they are capable to drive autonomous in both regions.
@@samuvisser It is the same approach for the learning part but Waymo is learning from way fewer cars in way fewer situations. If they had as much data as Tesla then they would need years to catch up after that point.
History is full of examples where the best technology doesn't always win. Only time will tell but my probability is on Tesla. If only the Simpsons could make an episode to alleviate all doubt. Good news is both determinations are only a function of time.
I think 99.9% people do not understand the relevance of this announcement for the automobile industry. Truly impressive
I don't think you understand your inflated sense of self.
@@gantmj He said something good about Tesla. Nothing about himself. He is right that this news should have been the biggest thing in years and yet people seem to be sleeping on it.
@@0ooTheMAXXoo0
He did say something about himself.
He sees himself as part of the top .1% of perceptive individuals, which in my opinion, is an inflated sense of self in this context.
gantmj What the hell are you talking about?
2:59:45
Elon: "all those lane charges have occurred with zero accidents"
Guy: "that is correct, I have watched ever single accident"
Elon looks back at him like "what?"
I think the software guy meant that accidents do happen (it's probably impossible to have 0% accident rate), but having watched most of the accident footage, the guy saw that none of these accidents happened during lane changes (could have happened at a traffic light or four-way intersection for example).
I still think the guy made a big mistake by saying that though; People should never talk with ambiguity in career defining presentations like this. Humans make mistakes of course but we need to try our best to avoid & learn from such mistakes.
Two years after first watching this, my mind still blows up with all the amazing tech and future vision discussed on this video. It's also wonderful to see the results we're now seeing in FSD 10.8 beta with more to come. This is truly fascinating and inspiring.
I watched the entire thing for first time in my professional life a conference 3+ hours long that is and this has been more exciting than a blockbuster movie. Elon you the man! Hats off to you Sir and to your vision and tenacity! Tenacity, tenacity, tenacity!
Was the first hour interesting? ;-)
This was less than 3 hours content.
I love how they are all so obviously engineers and it shows onstage.
I never understood why Tesla doesn't make a station wagon version
@@rosenbongurow3977 Not a big enough market for them, wouldn’t make sense for them to invest in…
can't wait to see the roadster in 2020 !
How much in the U.K ? Can Tesla car pull caravan ? How many miles to one charge ? Do you have change the batteries after 5 years ?