@@underratedblastoise3908 sure but I’m imagining grok being a tour guide of sorts. Discussing interesting and useful facts while answering questions. Road trips would be an amazing learning experience.
Dillon as always (but especially today) YOU ARE GREAT!!! Nobody else has gone into depth (in a short, understandable way) re. simulation vs. real world data as well as you have. I appreciate the time it took to research this to make this solid video. THANK YOU!!!
Gary Black should take a look at Google's dominance with search engines. FSD will also be a winner takes most market and it will be even harder to compete. And still no one have beaten Google at their game.
@@BrunoHeggli-zp3nl "They already do!" So the other OEMs have their own FSD system ready to be launched within a year? Or did they instead scrap their investments into FSD?
Mark my word… Some legacy OEMs will spend time & billions for NVIDIA's FSD solution but, if they are not bankrupt, endup licensing Tesla's solution in the end.
The whole point of going down the Nvidia path is to reduce the risk and cost for auto makers. Other automakers using Nvidia would only be paying if they go with such a system. I would assume they would only do so if they thought the system was good.
Great explanation of the difference between synthetic data and real world data. I think the comparison of the "Telephone" game is a great way to convey the difference to laymen.
I recently liquidated some long-term positions and now have about $250k to invest. Do you think Nvidia is still a smart buy, or have I missed a key buying window? Any recommendations for high-performing stocks would be appreciated.
I used to question the value of having a financial advisor until my wife’s company assigned her one in 2020. Honestly, it’s been the best financial decision I’ve made.
I consistently recommend Rebecca Lynne Buie as my top choice. She is well-known for her expertise in financial markets and has an impressive track record. I highly endorse her services.
Thanks for your talented work. I feel I don't have to watch the NVIDIA presentation now as your cogent, sophisticated analysis has given me what I need to feel certainty regarding Tesla's lead.
Thanks Dylan for this video. The synthetic data segment is a great reminder of why NVIDIA has not successfully created a fully autonomous vehicle despite all of their computing power. :)
OEMs all have teams to train models. None are starting at zero. The NVDA tools will be more helpful then u are suggesting IMHO. Thx for the video. I enjoyed watching
Thanks for another great video with excellent insight on the Autonomous Driving race. And please do let us know your thoughts on the general markets, as that's much appreciated also!
As an ai researcher, I am quite impressed about your research about the synthesis data issues. And yes, synthesis data has the mode collapse issue such that it cannot generate sufficient edge cases compared to real world data. But I want to add one more point about nvidia 3d world simulator. Even in the common cases, you still need to simulate the human behaviour of other cars for interactions. If they can simulate that, they have already solved the fsd. On the other hands, I can see that this nvidia 3d simulator will further improve tesla fsd and optimus development as I think enforcing the model to learn physic explicitly may be a good additional step. Nonetheless, ai field is moving extremely fast. May be after one year, the previous theory may not be applicable. We have seen the argument about the necessity of lidar in fsd in the past. We may see the argument about the necessity of real data in the future.
As a retired autopilot designer, I am well aware of the differences between simulation and the real world. Simulation can get your design into the ball park, but you need real world data to get into your seat! It is very challenging to predict what humans are going to do in any situation.
I have a free FSD, it is so much better than the other free months I was given. I trust it way more and have not had one disengagement for safety. 😊😊❤❤
@@MegaWilderness Well Panasonic North America is not in one of those overseas countries, so it is a bit silly of them to call Canada “overseas” from their operations here.
Try living in Alaska. Many, many businesses won't ship there, even if you pay for shipping! Alibaba will ship to Nowheresville, Africa, but won't ship to Anchorage. 🤬
This release & huge exposure of AV components by NVIDIA will certainly energise All the AV hopefuls in the Auto Industry BUT hopefully, when they truly and deeply evaluate this 3rd Party Tool kit they will realise that Licensing the entire suite from Tesla (Including the real world data) is the ONLY way they genuinely compete . Thanks Jensen .
Dylan, do you drink more than one serving of AG1 per day? I’ve just started using it, but I was curious because you talk about sipping all during the day and even into the evening and it felt like you wouldn’t be able to really do that with just one serving?
My compliments for your very solid analysis on FSD and the low impact of the new NVDIA synthetic products on the market ratios. Actually I think it is a bit mean of NVDIA to give the legacy carmakers hope, while the smart people of the world have known for a long time that they have absolutely no chance to complete this by themselves. You don't hold a lollipop in front of a child with diabetes, do you? The only chance they have left: 1. develop a car as soon as possible with a. the Tesla ICT architecture (6 computers instead of 150) b. the cheap production techniques of Tesla (casting etc.) and c. the Tesla camera set so that they can 2. license Tesla's FSD Point 1. above is hard enough for them.
Tesla vs the world is the same as Linux vs windows. Open source is people working together. Saying no one will ever catch tesla is short sighted. I am rooting for the open source version from the likes of NVIDIA. We all know that in the near future, when the ony real limiting factor battery range is solved, it will be back to features and style. Leveling the field again. Its all about the batteries.
Synthetic data is good for getting started with a project and to make a simple prototype, but it's no replacement for real world data if you actually want to release a product.
@@MegaWilderness No, that's not true. You'll always want to be working on the next version. Otherwise why bother with an iphone 10, 11, 12,13,14,15,16,etc. I was questioning the computer failures in the HW4 system that some people appear to be experiencing. I'm curious how widespread it is and what the fix may be. My '21 MYLR (HW3) was T-boned in early December and still in the shop. Looks like they won't be totaling it, but I'm still going to swap it out for a Juniper Y LR. It won't have AI5 until late this year, so I'll be getting an AI4 computer.
@@DillonLoomis22 Hey Dillon. What I've read in some articles is that some AI4 computers are glitching. Random functions seem to stop working and can end up bricking the car. This appears to be happening right out the door to several hundred miles down the road. The thing is, I haven't heard about this from any, what I call reputable, sources in the community. The article I read was from Jalopnik, which has a major anti-Tesla bias. Was wondering if you'd heard anything bubbling up from the group.
Will a Tesla data connectivity plan (with Grok) include a SpaceX LTE data (&voice) cellular plan? Are the current cellular aerials on a Tesla in need of any change to further improve SpaceX voice and data links!
Jensen says Tesla's data centers, a.i. algorithms, and massive real-world data give it a massive lead, plus, Tesla's focus upon developing autonomous vehicles, autonomous robots, and AGI are the three most important objectives and Elon is dedicated to solving all of them. Elon and Tesla are focused and they are winning.
IMHO, the synthetic is fine, but you must populate real world data to supplement in a large measure. This still keeps Tesla ahead for some time. Hopefully others will not go belly up before they can get real products on the road.
I just went through a time portal into the future. 10 days ago I had never used FSD nor have I used any “voice commands” of any kind anywhere…I don’t use Siri, Alexa or any other female assistant with all the knowledge of Google Search. Now, I use FSD for most of my driving..and it sounds like I’ll be talking to some Tolkien monster named Grok that has all the knowledge of the universe to assist me while my car drives me around…
The Chinese government will help make sure Chinese OEMs have access to autonomous driving capabilities. This may involve collaborative development of generic Chinese world models, training systems and AI driver models using Nvidia tools with companies like Huawei, Xiaomi, BYD, SAIC, Geely/Zeekr, Nio, Xpeng and others. I think they will be very fast followers, and in some cases possibly ahead, of Tesla’s Robotaxi deployments.
AI Autonomy training and Nano Computing seem to currently have a similar challenge -- double checking the data as you go along. (If only there were an intelligence there to do that, in real time.) But the Grok inclusion sounds fun.
Really? Why on Earth would most viewers not want macro market analysis? I find it extremely helpful in understanding this Wall Street sandbox in which we play. Thanks for including it. Stuff like this (10 year treasury bonds, etc.) is precisely the kind of macro indicators that Wall Street reacts to that individuals, such as myself, are ignorant of. As always, thanks!
when i used to talk about it more i'd get plenty of comments about "only here for tesla" and "not interested in macro" lol - i will include things as i find them relevant briefly tho :)
Even FSD with the most real world data has issues that I find on daily basis. Today I was coming up to a turn with ice on the road right before the turn and car wasn’t accounting for the low traction on ice and was going faster then comfortable so I had to take over and veer a bit to the left to avoid the ice patch and make the turn safely. Synthetic data is gong to be garbage 😂😂😂
I disagree. I think it is winner take all, and the reason is safety. Tesla has an insurmountable lead in real world data that is getting larger by the day. No one will switch to a different system once Tesla establishes safety dominance.
@@jebes909090 That mirage just drove me 20 miles through soul crushing Phoenix traffic to my hotel. I did nothing except listen to music and enjoy the ride. One hell of a mirage I’d say.
"I disagree. I think it is winner take all, and the reason is safety. Tesla has an insurmountable lead in real world data that is getting larger by the day. No one will switch to a different system once Tesla establishes safety dominance." That's correct. It's like how no one switching from Google's search engines to a competitor. Only this time the price is your own survival.
Regarding ASS, as usual Tesla is moving much faster than the regulators who are investigating incidents that occurred with older versions of the software and likely no longer apply. However, I will be shocked if NTSA doesn't raise concerns with the front bumper blind spot. At a minimum they will require users to have line of sight of the vehicle before engaging ASS which really makes the feature just a party trick. When Tesla first released ASS they made the car first back up 2 or 3 feet before going forward. Telsa knows the front bumper blind spot is an issue. An issue that should have been addressed a long time ago and is getting/will be addressed with upgraded hardware. The new model Y and the Cybercab reportedly will have a front bumper camera. But that doesn't immediately solve the problem because that extra bit of data has to be integrated into the software. I guarantee you it is going to be a while before they implement ASS on the Cybertruck because they can't have that thing driving around without a driver and without including data from the front nose camera. All these different branches of the software, because of significant hardware variances, is becoming and will increasingly become unmanageable for Tesla from the perspective of feature complete and especially driverless / unsupervised FSD & ASS. They will have to standardize on a miminum hardware requirement for an unsupervised feature set. They haven't even put that miminum hardware set into every new car yet, let alone the existing fleet. FSD has gotten very good but unsupervised and driverless are exponentially harder to achieve. There are reasons they are designing AI5.
As for Cosmo: It seems to me that all the worlds companies sharing this photoreal physics based info is good for all of us. All the data is common to all and exponentially grown. Everyone knows the crowd always outsmarts the one. Its not all about making money on Tesla stock. This is move for HUMANITY not Tesla stock owners. Picking a team like this is bad for all of us.
Also, FSD is fascinating. But after spending time in a relatives brand new Model Y, all passengers were stressed out while the car drove and I felt, even though it is OK, the threat of a computer glitching is too much of a risk. I mean how many times have you restarted your phone in the last year? I really don't need to be driven 99.9 % of the time. And wll never own a car without a steering wheel. My cars maps have gotten lost so many times I lost count.
Tesla will have 5 -10 years of monopoly on FSD. The key will be starting in the correct cities and scaling very fast. Orlando is the simplest city in the world to drive in.
I remember the chaos in the cell phone marketplace when Apple released the smartphone. Google released Android to overcome Apple's dominance. My guess is Google once again will release some FSD SDK, maybe in cooperation with Nvidia, to overcome Tesla's dominance. Google will take Waymo's FSD technology plus real world driving data to develop a solution. They will leverage their AI expertise. It will not catch up to Tesla but may provide an acceptable solution.
The design of Hondas new vehicles promotes chain pileups on roads- Whatever happened to the vehicles glass house to enable a line of sight beyond the vehicle in front
So at best Legacy auto and other pure play EV start-ups are 10 years behind where Tesla is now with regards to autonomy with Nvidia's help.I actually think that there is more opertunity for humanoid Robotics from other players besides Tesla ,though I still see Tesla as the leader here.At least in the sphere of Humanoid Robots the market is so vast(in the billions of units) that it will take many years for it to be satisfied.That's not the case for cars which feels like a shrinking market with the peak of 100 million new units per year having been reached a few years ago.
I'm someone who believes that the version of performative DEI that has been adopted by much of corporate America is flawed. Having said that, how can it be that diversity of inputs is essential for training AI to be most effective for dealing with real world scenarios and yet somehow not worth having when it comes to corporate management?
If you are going to continue to peddle AG-1 products with a 'clinical trial' as a justification, please include the reference to the published trial data (ie, the journal information about the published data/trial results). There are so many pseudo-science 'publications' out there at this point, that it really does make a difference to know what journal published the trial data/results. Looking for the actual credentials of the journal will help in understanding the credibility of the published data/results.
Cool that he's playing POE 2, I have over 200 hours now. The only way Jensen could be a "problem" for Tesla is if he stops selling Musk his GPU's. Non issue. Thanks.
NVIDIA is trying to jump start 'everyone else' following Tesla's lead. They are making the tools as you said but that is not even the start. Just as Tesla has been doing OTA updates (a trivial comparison with FSD) for a dozen years, the legacy automakers to this day cannot routinely update OTA. Automakers will first have to grow a software competency, then develop the software defined vehicle, then produce a fleet of vehicles using NVIDIA's tools, then march up the nines. NVIDIA saved a few steps... Minimum of a decade behind... possibly infinitely behind.
I still don’t understand why you and others are applauding that Tesla uses a third party AI. Tesla could’ve done this in-house but Elon chose to do it in another venture that none of us got to invest in. We got partially robbed 🤷♀️
@@grantzilla "@Steph1 Tesla is a mature company." No. It's not. "He could not attract Ai expertise by hiring via Tesla. " Nonsense. What kind of expertise do you believe is working on FSD? Janitors? "XAi was needed to allow talent to get in at ground level." No. it was not. There are other reasons for Elon's choice.
As far as FSD competition is concerned, Gary Black continues to get it wrong. Autonomy will be achieved with a combination of HW and SOFTWARE. On the software side it will be a vehicle (robot) OPERATING SYSTEM. There has never been many operating systems at any given time. There has always been just 2 primary operating systems for the devices and the hardware for any given period (Windows/Mac, Android/IOS, Linux/MS Server, etc). With all those historical examples the hardware and the software both evolve over time but is standardized for a given period. AI operated robots will be no different but there won't be a dozen suppliers for the operating system, there will only be 1 or 2. Which I don't think is that much different than much of the ADAS systems automakers use today. They all use the same few systems provided by tier 1 suppliers. Which I think is partly Gary's point but who will provide the autonomous vehicle operating system of the future to the automakers that survive the EV and autonomous transition? Seems clear Tesla is in the lead to create that operating system. Humanoid robots will be no different, 1 or 2 major players will provide the operating system for a standardized hardware set that will evolve over time (and eventually get displaced with something new). Everyone is focused on all the players trying to build humanoid robots (the hardware) but the big winner (by profit margin) is going to be the provider of the robot OPERATING SYSTEM and the entire ecosystem it will create. NVIDIA is simply capitalizing on the race to create the next generation AI based operating system for autonomous machines.
Grok should be connected to Tesla cameras so it can comment on landmarks. Be so great for road trips.
Good idea. 💡
Oh wow it just keeps getting better! Nice idea!
A map does it
It would cost around $3/s in compute
@@underratedblastoise3908
sure but I’m imagining grok being a tour guide of sorts. Discussing interesting and useful facts while answering questions. Road trips would be an amazing learning experience.
Dillon as always (but especially today) YOU ARE GREAT!!! Nobody else has gone into depth (in a short, understandable way) re. simulation vs. real world data as well as you have. I appreciate the time it took to research this to make this solid video. THANK YOU!!!
Excellent arguments regarding synthetic data. Thanks so much for your research and work!!
Thanks!
very very generous allan...thank you!!
Very thorough explanation on the synthetic/real world dichotomy, thank you Dylan
Barry Glack believes that the other OEMs will be able to find the “catch up with Tesla lever and switch it on”
They already do!
Gary Black should take a look at Google's dominance with search engines. FSD will also be a winner takes most market and it will be even harder to compete. And still no one have beaten Google at their game.
@@BrunoHeggli-zp3nl "They already do!"
So the other OEMs have their own FSD system ready to be launched within a year? Or did they instead scrap their investments into FSD?
😂😂😂😂
O p
Great job of explaining the synthetic data,Dillion!!
Great update Dillon. Many thanks
Great episode. Thx for the context and research.
Glad that NITSA is making sure that all manufacturers' cars are safe.
Mark my word… Some legacy OEMs will spend time & billions for NVIDIA's FSD solution but, if they are not bankrupt, endup licensing Tesla's solution in the end.
ouch, truth hurts :)
Delusional!
@@BrunoHeggli-zp3nl "Delusional!"
That's actually a great description of legacy auto.
The whole point of going down the Nvidia path is to reduce the risk and cost for auto makers. Other automakers using Nvidia would only be paying if they go with such a system. I would assume they would only do so if they thought the system was good.
1:14 let there be light!
Dillon of course I do want your analyses of the macro market ! This program was also very informative re AI and FSD, thank you. 😊
Excellent explanation about the limitations of synthetic data. 👍
thank you, i hope it was helpful!
Great explanation of the difference between synthetic data and real world data. I think the comparison of the "Telephone" game is a great way to convey the difference to laymen.
Great, as usual, thanks Dillon!! You have a nice day as well!
thanks for explaining all of this !!!
I NEED to know how the macro affects $TSLA. Please keep including this when appropriate. 🙏🏻
Dropped by to add this in too.
I recently liquidated some long-term positions and now have about $250k to invest. Do you think Nvidia is still a smart buy, or have I missed a key buying window? Any recommendations for high-performing stocks would be appreciated.
In my view, Palantir, ARM, and Nvidia are still strong investment opportunities-but to be clear, I'm not a financial advisor!
I used to question the value of having a financial advisor until my wife’s company assigned her one in 2020. Honestly, it’s been the best financial decision I’ve made.
How can I find a trusted financial planner like yours?
I consistently recommend Rebecca Lynne Buie as my top choice. She is well-known for her expertise in financial markets and has an impressive track record. I highly endorse her services.
Thank you for the suggestion. I found her website and filled out the inquiry form. I’m hoping to hear back from her soon.
Very thorough video! Thanks, Dillon!
Thanks for your talented work. I feel I don't have to watch the NVIDIA presentation now as your cogent, sophisticated analysis has given me what I need to feel certainty regarding Tesla's lead.
Thanks Dylan for this video. The synthetic data segment is a great reminder of why NVIDIA has not successfully created a fully autonomous vehicle despite all of their computing power. :)
awesome! One of your more interesting videos for those into AI and who watched the keynote last night. Thanks for processing feedback from others..
OEMs all have teams to train models. None are starting at zero. The NVDA tools will be more helpful then u are suggesting IMHO. Thx for the video. I enjoyed watching
Good Job on the synthetic data! Also - I signed up for AG1 :)
thank you for supporting the channel in that way...it does help way more than most people realize. hope you enjoy it barry :)
Very well explained (synthetic data, real data)!
Thanks for another great video with excellent insight on the Autonomous Driving race.
And please do let us know your thoughts on the general markets, as that's much appreciated also!
As an ai researcher, I am quite impressed about your research about the synthesis data issues. And yes, synthesis data has the mode collapse issue such that it cannot generate sufficient edge cases compared to real world data. But I want to add one more point about nvidia 3d world simulator. Even in the common cases, you still need to simulate the human behaviour of other cars for interactions. If they can simulate that, they have already solved the fsd. On the other hands, I can see that this nvidia 3d simulator will further improve tesla fsd and optimus development as I think enforcing the model to learn physic explicitly may be a good additional step.
Nonetheless, ai field is moving extremely fast. May be after one year, the previous theory may not be applicable. We have seen the argument about the necessity of lidar in fsd in the past. We may see the argument about the necessity of real data in the future.
12 complaints out of 7+ million cars... that is a waste of resources.
As a retired autopilot designer, I am well aware of the differences between simulation and the real world. Simulation can get your design into the ball park, but you need real world data to get into your seat! It is very challenging to predict what humans are going to do in any situation.
So, Nvidia is selling scalpels but your going to need to find people with specialized skills if you need brain surgery.
EVry LEGA CEO is neuroactively challenged. 🧠❌
basically this right here :)
I have a free FSD, it is so much better than the other free months I was given. I trust it way more and have not had one disengagement for safety. 😊😊❤❤
Thanks Dillon well explained
Excellent explanation and coverage on real world data vs synthetic data for AV training. Tesla is years ahead.
But how can you solve autonomy if you are bankrupt?
...to what do you refer?
Interesting that some reporters think Canada is “overseas.” 😂
24:30 Yeah, what sea do you have to go over to get from the US to Canada, Panasonic?
Lake Ontario is pretty big…
It is for most countries, get outside your myopic bubble
@@MegaWilderness Well Panasonic North America is not in one of those overseas countries, so it is a bit silly of them to call Canada “overseas” from their operations here.
Try living in Alaska. Many, many businesses won't ship there, even if you pay for shipping! Alibaba will ship to Nowheresville, Africa, but won't ship to Anchorage. 🤬
Wow, what a well done video. When are the Tesla RUclips awards?
Good evening Dillon!! Anxious to hear tonight’s report!! 😎👍 Thank You!!
Good stuff. Reminder Macro market represents a "group-think" not a consciousness.
Great episode!
This release & huge exposure of AV components by NVIDIA will certainly energise All the AV hopefuls in the Auto Industry BUT hopefully, when they truly and deeply evaluate this 3rd Party Tool kit they will realise that Licensing the entire suite from Tesla (Including the real world data) is the ONLY way they genuinely compete . Thanks Jensen .
very well done I think that was a great summary of synthetic data.
Dylan, do you drink more than one serving of AG1 per day? I’ve just started using it, but I was curious because you talk about sipping all during the day and even into the evening and it felt like you wouldn’t be able to really do that with just one serving?
just one serving...im a slow drinker :)
thank you for checking it out...hope you enjoy it!
My compliments for your very solid analysis on FSD and the low impact of the new NVDIA synthetic products on the market ratios.
Actually I think it is a bit mean of NVDIA to give the legacy carmakers hope, while the smart people of the world have known for a long time that they have absolutely no chance to complete this by themselves. You don't hold a lollipop in front of a child with diabetes, do you?
The only chance they have left:
1. develop a car as soon as possible with
a. the Tesla ICT architecture (6 computers instead of 150)
b. the cheap production techniques of Tesla (casting etc.) and
c. the Tesla camera set so that they can
2. license Tesla's FSD
Point 1. above is hard enough for them.
Tesla vs the world is the same as Linux vs windows. Open source is people working together. Saying no one will ever catch tesla is short sighted. I am rooting for the open source version from the likes of NVIDIA. We all know that in the near future, when the ony real limiting factor battery range is solved, it will be back to features and style. Leveling the field again. Its all about the batteries.
Nvidia will give legacy automakers an opportunity to develop autonomous systems for hybrids and ICE vehicles. This is my main takeaway.
Grok + Tesla = 🔥🔥🔥🔥🔥🔥🔥🔥
Is SoundHound involved?
Synthetic data is good for getting started with a project and to make a simple prototype, but it's no replacement for real world data if you actually want to release a product.
I recently heard about the AI4 hardware issue. Any thoughts about it?
No
If it was good enough they wouldn't be working on AI5
@@MegaWilderness No, that's not true. You'll always want to be working on the next version. Otherwise why bother with an iphone 10, 11, 12,13,14,15,16,etc. I was questioning the computer failures in the HW4 system that some people appear to be experiencing. I'm curious how widespread it is and what the fix may be. My '21 MYLR (HW3) was T-boned in early December and still in the shop. Looks like they won't be totaling it, but I'm still going to swap it out for a Juniper Y LR. It won't have AI5 until late this year, so I'll be getting an AI4 computer.
which issue are you referring to?
@@DillonLoomis22 Hey Dillon. What I've read in some articles is that some AI4 computers are glitching. Random functions seem to stop working and can end up bricking the car. This appears to be happening right out the door to several hundred miles down the road. The thing is, I haven't heard about this from any, what I call reputable, sources in the community. The article I read was from Jalopnik, which has a major anti-Tesla bias. Was wondering if you'd heard anything bubbling up from the group.
Synthetic data is the pinky finger. OEM’s still need: a wrist, a palm, a thumb, an index finger, a middle finger and a ring finger.
No, they only get the middle finger...
Will a Tesla data connectivity plan (with Grok) include a SpaceX LTE data (&voice) cellular plan? Are the current cellular aerials on a Tesla in need of any change to further improve SpaceX voice and data links!
Jensen says Tesla's data centers, a.i. algorithms, and massive real-world data give it a massive lead, plus, Tesla's focus upon developing autonomous vehicles, autonomous robots, and AGI are the three most important objectives and Elon is dedicated to solving all of them. Elon and Tesla are focused and they are winning.
IMHO, the synthetic is fine, but you must populate real world data to supplement in a large measure. This still keeps Tesla ahead for some time. Hopefully others will not go belly up before they can get real products on the road.
Using only synthetic data is like treating the real world as a car game driving in fantasy lands, it will not be safe enough imo.
Are you telling me Grand Theft Auto is FAKE?!!
I just went through a time portal into the future.
10 days ago I had never used FSD nor have I used any “voice commands” of any kind anywhere…I don’t use Siri, Alexa or any other female assistant with all the knowledge of Google Search.
Now, I use FSD for most of my driving..and it sounds like I’ll be talking to some Tolkien monster named Grok that has all the knowledge of the universe to assist me while my car drives me around…
Tools. They're all tools.
Same here, for a month.
The Chinese government will help make sure Chinese OEMs have access to autonomous driving capabilities. This may involve collaborative development of generic Chinese world models, training systems and AI driver models using Nvidia tools with companies like Huawei, Xiaomi, BYD, SAIC, Geely/Zeekr, Nio, Xpeng and others. I think they will be very fast followers, and in some cases possibly ahead, of Tesla’s Robotaxi deployments.
Next level video.
End story points to Tesla technology is still the very best and the most reliable ❤
AI Autonomy training and Nano Computing seem to currently have a similar challenge -- double checking the data as you go along. (If only there were an intelligence there to do that, in real time.)
But the Grok inclusion sounds fun.
9:27 on that winter bulk? 💪🏼
def carrying a few extra lbs thru this winter arc lol :)
5:50 What did BMW do in 2019?
Sounds like the market is pricing in possible upcoming tariffs.
Really? Why on Earth would most viewers not want macro market analysis? I find it extremely helpful in understanding this Wall Street sandbox in which we play. Thanks for including it. Stuff like this (10 year treasury bonds, etc.) is precisely the kind of macro indicators that Wall Street reacts to that individuals, such as myself, are ignorant of. As always, thanks!
when i used to talk about it more i'd get plenty of comments about "only here for tesla" and "not interested in macro" lol - i will include things as i find them relevant briefly tho :)
Even FSD with the most real world data has issues that I find on daily basis. Today I was coming up to a turn with ice on the road right before the turn and car wasn’t accounting for the low traction on ice and was going faster then comfortable so I had to take over and veer a bit to the left to avoid the ice patch and make the turn safely. Synthetic data is gong to be garbage 😂😂😂
Yes I need a car I can talk to too distracting to type on a screen
Excellent video
Grok should be able to do work in the car and save it to the cloud.
Dylan and all, heads up. The China Tesla site is now updated with the new Model Y.
Dillon - what does "solved" mean? I mean when or how do you measure "solved"?
Most people would say solved is better than human level driving
hey dan - i'd say when tesla shifts to unsupervised...that's the milestone for me
@@DillonLoomis22 I heard Elon use the terms "safer than the average driver" yesterday.
I disagree. I think it is winner take all, and the reason is safety. Tesla has an insurmountable lead in real world data that is getting larger by the day. No one will switch to a different system once Tesla establishes safety dominance.
Tesla will never get there. Fsd is basically a mirage.
@@jebes909090 That mirage just drove me 20 miles through soul crushing Phoenix traffic to my hotel. I did nothing except listen to music and enjoy the ride. One hell of a mirage I’d say.
@@alexedens1 until it crashes you one day unexpectedly. thats the problem, the closer it gets, the more unexpectedly i will fail.
"I disagree. I think it is winner take all, and the reason is safety. Tesla has an insurmountable lead in real world data that is getting larger by the day. No one will switch to a different system once Tesla establishes safety dominance."
That's correct. It's like how no one switching from Google's search engines to a competitor. Only this time the price is your own survival.
@@jebes909090 "Tesla will never get there. Fsd is basically a mirage."
So stop fantasizing and look at the reality instead?
Regarding ASS, as usual Tesla is moving much faster than the regulators who are investigating incidents that occurred with older versions of the software and likely no longer apply. However, I will be shocked if NTSA doesn't raise concerns with the front bumper blind spot. At a minimum they will require users to have line of sight of the vehicle before engaging ASS which really makes the feature just a party trick. When Tesla first released ASS they made the car first back up 2 or 3 feet before going forward. Telsa knows the front bumper blind spot is an issue. An issue that should have been addressed a long time ago and is getting/will be addressed with upgraded hardware. The new model Y and the Cybercab reportedly will have a front bumper camera. But that doesn't immediately solve the problem because that extra bit of data has to be integrated into the software. I guarantee you it is going to be a while before they implement ASS on the Cybertruck because they can't have that thing driving around without a driver and without including data from the front nose camera. All these different branches of the software, because of significant hardware variances, is becoming and will increasingly become unmanageable for Tesla from the perspective of feature complete and especially driverless / unsupervised FSD & ASS. They will have to standardize on a miminum hardware requirement for an unsupervised feature set. They haven't even put that miminum hardware set into every new car yet, let alone the existing fleet. FSD has gotten very good but unsupervised and driverless are exponentially harder to achieve. There are reasons they are designing AI5.
As for Cosmo: It seems to me that all the worlds companies sharing this photoreal physics based info is good for all of us. All the data is common to all and exponentially grown. Everyone knows the crowd always outsmarts the one. Its not all about making money on Tesla stock. This is move for HUMANITY not Tesla stock owners. Picking a team like this is bad for all of us.
Still my favorite Tesla You Tube analyst alway makes the complex easy to understand
Also, FSD is fascinating. But after spending time in a relatives brand new Model Y, all passengers were stressed out while the car drove and I felt, even though it is OK, the threat of a computer glitching is too much of a risk. I mean how many times have you restarted your phone in the last year? I really don't need to be driven 99.9 % of the time. And wll never own a car without a steering wheel. My cars maps have gotten lost so many times I lost count.
👍🏼
Tesla will have 5 -10 years of monopoly on FSD.
The key will be starting in the correct cities and scaling very fast.
Orlando is the simplest city in the world to drive in.
, When is the hardware three upgrade supposed to happen? When is the update coming?
anyone's guess but if i had to guess i'd say not for another year or two
I remember the chaos in the cell phone marketplace when Apple released the smartphone. Google released Android to overcome Apple's dominance. My guess is Google once again will release some FSD SDK, maybe in cooperation with Nvidia, to overcome Tesla's dominance. Google will take Waymo's FSD technology plus real world driving data to develop a solution. They will leverage their AI expertise. It will not catch up to Tesla but may provide an acceptable solution.
The design of Hondas new vehicles promotes chain pileups on roads- Whatever happened to the vehicles glass house to enable a line of sight beyond the vehicle in front
Legacy OEMs can’t even build a freaking audio interface that works. These same “engineers” are supposed to create working autonomy?
So at best Legacy auto and other pure play EV start-ups are 10 years behind where Tesla is now with regards to autonomy with Nvidia's help.I actually think that there is more opertunity for humanoid Robotics from other players besides Tesla ,though I still see Tesla as the leader here.At least in the sphere of Humanoid Robots the market is so vast(in the billions of units) that it will take many years for it to be satisfied.That's not the case for cars which feels like a shrinking market with the peak of 100 million new units per year having been reached a few years ago.
Above 4.8% in the 10 yr bond is the danger zone for the stock market.
definitely something i think more investors should be aware of
👋 SOMEBODY 🙏 please ask GARY what he is "SMOKING..." ⁉ 🤣
I'm someone who believes that the version of performative DEI that has been adopted by much of corporate America is flawed. Having said that, how can it be that diversity of inputs is essential for training AI to be most effective for dealing with real world scenarios and yet somehow not worth having when it comes to corporate management?
Diversity of viewpoints, outliers, etc. can't get that from homogenized inputs
If you are going to continue to peddle AG-1 products with a 'clinical trial' as a justification, please include the reference to the published trial data (ie, the journal information about the published data/trial results). There are so many pseudo-science 'publications' out there at this point, that it really does make a difference to know what journal published the trial data/results. Looking for the actual credentials of the journal will help in understanding the credibility of the published data/results.
Cool that he's playing POE 2, I have over 200 hours now. The only way Jensen could be a "problem" for Tesla is if he stops selling Musk his GPU's. Non issue. Thanks.
NVIDIA is trying to jump start 'everyone else' following Tesla's lead. They are making the tools as you said but that is not even the start. Just as Tesla has been doing OTA updates (a trivial comparison with FSD) for a dozen years, the legacy automakers to this day cannot routinely update OTA. Automakers will first have to grow a software competency, then develop the software defined vehicle, then produce a fleet of vehicles using NVIDIA's tools, then march up the nines. NVIDIA saved a few steps...
Minimum of a decade behind... possibly infinitely behind.
What if Tesla sells a Grok API to other automakers? Instead of the dreadful voice command systems other companies have
I hope it improves voice commands work under GROK because our Tesla doesn't understand the Aussie accent so useless 😢.
Thank goodness they will have groq in the car. Tesla voice integration is terrible at the moment.
사이버트럭 가격 다운은 언제 된데요? 지금 8만이라던데..6만까지는 내려 와야..
I still don’t understand why you and others are applauding that Tesla uses a third party AI. Tesla could’ve done this in-house but Elon chose to do it in another venture that none of us got to invest in. We got partially robbed 🤷♀️
What are you talking about? Grok is an LLM.
@ made by Xai which Tesla has no stake in, and will end up paying licensing fees to even though it could’ve been developed in-house at Tesla
@@Steph1Tesla is a mature company. He could not attract Ai expertise by hiring via Tesla. XAi was needed to allow talent to get in at ground level.
@@viewer_5714 "What are you talking about? Grok is an LLM."
He is clearly arguing that Tesla should have made Grok inhouse instead.
@@grantzilla "@Steph1 Tesla is a mature company."
No. It's not.
"He could not attract Ai expertise by hiring via Tesla. "
Nonsense. What kind of expertise do you believe is working on FSD? Janitors?
"XAi was needed to allow talent to get in at ground level."
No. it was not. There are other reasons for Elon's choice.
Training on synthetic data will create an AI centipede.
Dang - U B good
Where is all the talk of LiDAR now?
AI takes megawatts, the Human Brain 20 watts!!
As far as FSD competition is concerned, Gary Black continues to get it wrong. Autonomy will be achieved with a combination of HW and SOFTWARE. On the software side it will be a vehicle (robot) OPERATING SYSTEM.
There has never been many operating systems at any given time. There has always been just 2 primary operating systems for the devices and the hardware for any given period (Windows/Mac, Android/IOS, Linux/MS Server, etc). With all those historical examples the hardware and the software both evolve over time but is standardized for a given period. AI operated robots will be no different but there won't be a dozen suppliers for the operating system, there will only be 1 or 2.
Which I don't think is that much different than much of the ADAS systems automakers use today. They all use the same few systems provided by tier 1 suppliers. Which I think is partly Gary's point but who will provide the autonomous vehicle operating system of the future to the automakers that survive the EV and autonomous transition? Seems clear Tesla is in the lead to create that operating system.
Humanoid robots will be no different, 1 or 2 major players will provide the operating system for a standardized hardware set that will evolve over time (and eventually get displaced with something new). Everyone is focused on all the players trying to build humanoid robots (the hardware) but the big winner (by profit margin) is going to be the provider of the robot OPERATING SYSTEM and the entire ecosystem it will create. NVIDIA is simply capitalizing on the race to create the next generation AI based operating system for autonomous machines.
I am sorry there is no such thing as a general world model.
Yes there is.
Everyone working on autonomous driving is using the same technique
Not really. Most others are going with a much more hardware rich set up.
@TheEvilmooseofdoom Tell me exactly what Wayve is doing. Now under Uber's wing