Tesla just revealed this SECRET FSD Advantage (Ep. 733)
HTML-код
- Опубликовано: 9 июн 2024
- How close is Tesla FSD to solving autonomous driving? How close is the competition? What are the challenges?
Tesla is planning to have 100 Exa-Flops of compute power to train FSD by end of next year. / 1673307285873983488
Social
🐦 Twitter / heydave7
🎧 Apple Podcast: podcasts.apple.com/us/podcast...
🎙️Spotify open.spotify.com/show/2iin015...
📸 Instagram / heydave7
Dave Lee on Parenting, / @daveleeonlife7659
NOTE: Please don’t fall for scammers who might impersonate me or others in the comment section. I do NOT give out my number or ask people to contact me.
💁🏻♂️ Disclaimer: All content on this channel is for discussion and illustrative purposes only and should not be construed as professional financial advice or recommendation to buy or sell any securities. Should you need such advice, consult a licensed financial or tax advisor. All views expressed are personal opinion as of date of recording and are subject to change without responsibility to update views. No guarantee is given regarding the accuracy of information on this channel. Neither host or guests can be held responsible for any direct or incidental loss incurred by applying any of the information offered. Author is long TSLA and other stocks at time of original video publish date.
Check out my archived articles/posts on Tesla and investing: teslamotorsclub.com/tmc/threa...
#Tesla #TSLA #Stocks
![[CVPR'23 WAD] Keynote - Ashok Elluswamy, Tesla](http://i.ytimg.com/vi/6x-Xb_uT7ts/mqdefault.jpg)
![[CVPR'23 WAD] Keynote - Ashok Elluswamy, Tesla](/img/tr.png)
Hope your enjoying your trip! We miss the more frequent videos.
Always love your reasonable, educated thoughts Dave!
Great points, Dave - and I totally agree with your analysis and concur with your timeline (but note that it's "iceberg" and not "glacier" for your analogy!) ...
yeah, I was considering sending a message to you Dave about iceberg versus glacier. I perfectly understood what you meant but readers who are not familiar with that analogy, say from another country, might not get it.Thanks for the great work that you do in processing information for us
@@MangoFrozen The "glacier" metaphor I substituted in my head with "undersea volcano". That worked far better for me.
Glacier? Don't you mean iceberg? 😂
0:25 biggest point 1st.. Thank you dave.
People don’t intuitively understand exponential growth, yet it exists everywhere in nature.
Small trajectory changes and inputs early on (ie decisions like going vision only years ago) yield massive strides over time. Love your short and sweet videos Dave!
for real life example, people should look into rats and feral pigs population explosion to get a general sense of how fast things can accumulate.
@@jasonk125 tesla does that too
@@jasonk125 Tesla AI days explained their data filtering--they throw away millions of events before it's even transmitted from the vehicle of uninteresting scenarios that are already well trained on.
As someone who has been watching multiple FSD vids every update since Jan 2021 shortly after the public beta started, I think it's pretty clear that exponential growth in data does not directly translate into exponential growth in FSD's quality. Not even getting into the sometimes significant regressions alongside major updates, the rate of improvement Jan 2021-2022 seemed dramatically faster than Jan 2022 to today. Which makes sense as it asymptotically approaches the hypothetical limit--but it's clear that it still far too frequently is making grave, potentially fatal, errors in addition to missing core necessary features for a fully autonomous vehicle (reading signs, U-turns, 3-point turns, still barely started on being responsive in driving to rain let alone snow/ice).
Fsd is very far from ready. It may beat the roadster.
Dave’s channel and content are a perfect example of quality over quantity.
Please come forward more often with your videos! We are missing you terribly.
Hope you're enjoying your travels, but always great to see you post a new vid! Regarding your analogy...I think you mean iceberg instead of glacier.
Dave, I like that you focus on whats fundamental, the big picture. Rather than what what Gordon Johnson said or every other tiny event that will be forgotten in a week.
I think you mean iceberg, not glacier
That's the first thing I caught, too ... :)
FSD beta is just the tip of the glacier
I agree and am working my ass off to stack as many shares of Tesla for the future as I can because I think the same thing.
Always great to see a Dave alert
Thanks Dave. This is informative. Would love to see more videos from your channel soon.
As a Tesla investor, the harder it is to solve FSD, the better it is for the stock once it is solved, because it will take A LONG time for anyone to then catch up.
As long as the competition is relying on LiDAR, even if it worked perfectly today they wouldn't be able to scale it up to volume robotaxis. The sensor suites used by others are simply too expensive ($100kish) and produced in volumes suitable to outfit maybe dozens of vehicles a year, it would take years to drive down costs and scale to outfitting millions of vehicles a year. AKA it already will take a LONG time for anyone to catch up.
Good point. A real possibility is that FSD is so good, it becomes the standard. Supporting this of course are all the other EV manufacturers joining the Tesla Charging standard. And part of this is Tesla openly sharing patents. FSD success and ubiquity would be hugely profitable, but for how long? (Definitely long enough for shareholders to profit, but 10 years? only 5 years?) Are there any scenarios where ubiquity results in reduced profit margins?
@@WillN2Go1 Tesla FSD will be the MS Windows.
That’s not how ai works. It’s mainly research driven.
Hi Dave and family - awesome to get a vid from you! Show us some of your travel experiences. Wonder if one would be able to take one’s Tesla along all over the world. One sure place to test FSD would be in developing countries where road rules are fluid
Very helpful and insightful Dave! Safe travels always.
Thanks Dave! Glad your enjoying your travels with family! 🌏
So good to hear you, relaxed , healthy and to the point.
Thank you for your ideas.
Wish you a save trip with your family.
😊
Dave, you gotta account for the journey and not just the destination.
Thank you for your insightful coverage as always David - from around the world 🙂
Brilliant points. Also, the stories of your travels are almost as fascinating as your business content! Thanks for posting!
Nice video! I think one challenge for FSD is to understand proper speed limits for the environment when no speed limit data is available or the data is wrong. In the space of 1 mile I deal with two instances where my Model Y thinks the speed limit is 55 but max safe speed for the area is 45 IMO and no speed limit signs are posted. It is a new housing development and office park. I also wonder if the accelerating influx of data to Tesla, combined with the increase in computing power will at some point figure it all out suddenly. We can dream:)
Your comments are so true and dead on accurate. Even drivers not active with fsd still gives data to Tesla.
Good video Dave, thanks for taking the time!
I follow you regularly Dave from your videos. Your optimism of FSD is contagious. Keep it coming. Hoping for a safer & better world's future via FSD.
Dave, I always love your perspective. Thanks for taking a break from your travels to share.
(I didn’t realize glaciers come up out of the water 😂)
Thank you so much for not going into the details of the trauma, this is why you are the Yoda of the Tesla community!!!;)
Wise and to the point:)
Enjoy your vacation! And thank you Dave!
i get sad when days or weeks go by and Dave Lee does not post a new video...hahahah... i appreciate all the info youve supplied us with over the years!
excellent - thanks Dave. I'm a side-line observer and am planning to purchase down the track a bit here in Australia. I guess what surprises me is that the FSD still seems to make rookie errors at this stage of its training. I realise autonomous driving is a very tough thing - but watching Chuck Cook's recent drives (including lurching over railway line crossings) leaves me a bit wondering if the team is stuck in a local maxima situation. I truly hope I'm wrong and that the exponential improvements will be visible shortly - the pace over the past few releases has seemed more incremental. Big city driving in Australia is going to be a LOT tougher than the many successful drives I watch (like Whole Mars)
Thanks for both your insights on Tesla and news on traveling with family!
Thanks Dave. Life will dramatically change in the future for the better with FSD once it is perfected. I am no spring chicken. At the age of 72 I will one day have to surrender my car do to being incompetent. I live in SF Calif a highly populated city. About 6 months ago my FSD took me from my house thru heavy traffic to get my appointment for a vaccination. I was so impressed that my car made the 4 mile trip with no intervention. I was a bit tense letting my car drive me to my destination but it did. Amazing is all I can say. I plan to visit my son in Los Angeles and can’t wait this week to use FSD. YES LIFE AS WE KNOW IT WON’t be the same. Maybe a few Tesla Bots to help me as I become less mobile. Heeeheeee. This won’t be my parents when they aged.
Most drivers in NJ are incompetent and they don't take their licenses away so if fsd doesn't work out you could move to NJ.
This sounds like a ChatGPT written comment. I call BS!
@@StA_V_ROSover 50% of comments on RUclips are spam anyways
I catch you on you tube mostly and follow you and the others as well. You have a technical approach I Enjoy.
You are blessed Dave and you are a blessing thank you enjoy your vacation
Thanks Dave
Great to hear from you !
Thanks for the beautiful analogy. I understand this a lot better.
Safe travels Dave for you and your family. I sure miss your videos.
Loved you video! THX !!!
Thank you Dave! We have missed you but happy to hear you are traveling with your family. I guess you heard about all the crazy fsd fud going on over here. I’m happy you gave your opinion. Have fun 😊
Thank you so much David lee for understanding better for future FSD will be happen. I am so exciting!
We need more from Dave !
Glaciers; Just FYI, Glaciers don't grow under water they are frozen rivers. When a piece of one falls in the ocean, it becomes a floating iceberg that immediately starts to melt. Maybe a volcano or seamount would be a better analogy for your description.
Thank you!
Boy if I could trade insecure SMR posting daily videos for Dave, I’d be a happy man ❤
SMR kinda can’t not watch them videos however I agree Dave’s view would be nice.
@@tasd5673 I’m sorry I’m not sure I understand your comment 🙂
They're on the same page.
Cheers Dave. I agree with your thinking but I think your glacier should be an iceberg
An iceberg is a piece of glacial ice floating in the ocean.
Good to see you again
Wow! Great to see another of your videos, Dave! I follow you because you're genuinely smart, integral and spiritual. I'm not sure if you believe in telepathy, but this morning, I suddenly wondered: "Oh, what would happen to Dave? I haven't seen any of his videos recently". And boom! I've just checked my computer and the first recommended video was this! Amazing, isn't it?
Magical thinking Bud
Thanks for sharing
Wow ! Thanks !
Agree with your take Dave.
FSD can always get “closer” to being solved without ever reaching perfection, e.g. causing no accidents, injuries or deaths. So at what threshold do regulators approve FSD, e.g. causing only one fatal accident per million trips? What would be the threshold?
FSD will provide hours of in car entertainment for future humans lol
So generous to offer your vast understanding and communication skills for the betterment of so much understanding. Thanks
PS like company’s that spin off new companies glaciers become a new entity when they “calve” off and become icebergs.
Regulatory approval!!! Biggest hurdle. Wait for change - 25 years. Some people in power still using an abacus.
First thanks for the frank discussion and, I want to thank the Tesla team for their hard work and getting us this far. I'm in the camp that we are a few years away from the target. Here in hilly, lumpy roads PA FSD is challenged, and to be fair so are human drivers. In many situations you can't see the cars hidden by the road depressions. You end up counting to 3 in hopes of pulling out to merge safely. Some other edge cases are not moving forward after a train gate has risen, but then speeding across the rough tracks. Not always recognizing the back of a semi tractor, when they aren't pulling a trailer. The most annoying item is considering side lanes that are only for business turnoffs as driving lanes. I partially solved this by reducing the lane changing action. On narrow mountain roads FSD works well generally, until you get to a bridge where the painted road line and bridge barrier are in the same vertical position, FSD drops out as you are getting too close to the barrier.
Overall, I've seen great progress over my 6 years of driving my MS!
Stay strong and take care
Fun video, but glaciers are built up of layers of snow falling on land and being compressed into ice. Then the weight causes the ice to flow down to the ocean, where it calves and becomes icebergs. Glaciers NEVER form underwater.
Warm regards, Rick.
Hi Dave, could you also share what are the lesson you learn at different places you visit? Take us on the journey with you if its possible. Thank you!
Dave , your analogy should be "Iceberg" , where 90% is below the surface :)
Yup, glaciers form on land.
Biggest improvement for FSD will be when people in charge of the roads start to cooperate
I love it!
Between the time the saga video came out until now, Tesla already fixed what went wrong and FSD stops at that stop line now. That blew my mind
The sign was well before the point where the car needed to stop. A weird edge case. Not really a fail but Ross felt he had to hit the brakes which was the right decision. Good Tesla fixed it so fast. Amazing.
Tesla FSD is impressively good and has been getting increasingly better especially with the last few updates. It’s much more confident in switching lanes, asserting itself in traffic, etc.
That said, on my daily route there are a couple of locations it has been struggling with and either always has or has regressed.
There’s a very sharp two lane left turn onto a highway on ramp. FSD fails to stay in its lane and would drive off other traffic taking the same turn at the same time if one doesn’t intervene.
Just up ahead is a another junction with two straight lanes and two left turn lanes. All lanes have lights. When waiting in the left turn lane FSD will occasionally apply the signals for going straight ahead to itself and try to drive out into the intersection while the left turn signal is still red.
Each of these is along my daily route and potentially catastrophic if not corrected for. FSD will need to reliably avoid failure like this *at scale* to be able to be driven unmonitored. I’m afraid it might still take quite a while before we are there.
There’s a road with a dedicated left turn lane that has a traffic light. FSD always mistakes the left turn traffic light to apply to the main forward direction and tries to stop when the left turn light is red going straight.
Good insight on what Tesla needs that compute for. Probably includes Optimus training.
Edge cases? You mean like the street sweeper that was traveling on the wrong side of the road in a cloud of dust as my car was trying to make a left turn? 😮
Thanks Dave! Visit Uzbekistan if you can. Turkey is awesome!
Stay safe, have fun!
Thank you very much for video Dave. You are right about competition only if the competition is taking the same route to solve the problem. If not Tesla's advantage plays no role...
Hey Dave, picking up on your point about Tesla using even the non-FSD vehicles to assist in training by gathering"edge" cases from the entire fleet, I'm trying to explain why my '22 Model Y with FSD continues to flub seemingly easy maneuvers in my quiet suburban neighborhood where dozens of folks own Teslas. I've been testing since 11.4.2 through 11.4.4 and I've seen no improvement in several specific cases. The most notable scenario is making a left turn at a basic 4-way intersection with no obstructions where only the cross street has a stop sign. No lane markings, wide streets, no parked cars near intersection. The expected behavior when no other traffic is present is that the car would slow a bit below the 25mph limit but NOT stop before making the turn. The Tesla seems confused every time as it approaches, and ends up hard braking as it comes to the edge of the intersection. I have intervened maybe a dozen times at this point across all these versions and recorded a message about how inappropriate it is to stop at such an intersection where there is no stop sign and no traffic and noted how dangerous it is because they guy BEHIND is not expecting to stop. I've intervened repeatedly for other situations as well that you'd think it should have learned from by now, but this is certainly the least excusable given the dead simplicity of the scenario. While I understand the theoretical data advantages and ML principles frequently discussed here, this real-life experience seems to contradict all that so curious what folks think might be going on. Thanks!
Keys at the very end!
Like the tip of an iceberg, indeed.... When the tipping point occurs with Dojo + massive real-world data from the fleet, the world as we know it changes. Then Optimus/Tesla Bots disrupting and redefining 'labor'... whoa Nellie!!! Hold on to your hats, because it's gonna be a wild ride!!! SoCalFreddy
I see the glacier!
we do not see Dave enough these days. great incisive analysis on what separates Tesla from other companies working on full self driving.
Uh….you mean an iceberg? A glacier is more like a frozen river I think.
Hey Dave
Have you sold at the peak and now traveling the world? Miss your videos and content. Enjoy life!
You should visit Greece. Beautiful !
Thank you Dave. Wonderful observations and knowledge! Isn't it great you can podcast from various countries while traveling. Ai is going to be crazy fabulous. Looking forward to having my robotaxi working for me, and my optimus helping do so many things. Seniors can take themselves out without relying on others all the time. Optimus can carry them, run errands. It will help people remain independent as they age.
This guy gets it 😉
Don't miss Greece and Italy from your tour, after Turkey😉
I was really hopeful of my investments this year, but I followed some stock suggestions that didn't go so well, I've been studying the stock market and I realized some investors made millions from the last recession and I was wondering if such success rate could be achieved in this present market. and the Federal Reserve taking a more hawkish approach to interest rates.
Yes, If you go long against the market you'd make such success.
I built a 7 figure well-diversified portfolio just by following my F.A Trisha Jean Webb's recommendations. I buy quality firms, anticipate to hold them regardless of what happens, pay up but not too much, keep track, sell only when necessary, and be ready to course correct. also ignore the forecasts and market views which are at best entertaining but completely useless.
@@RandyPelletier I did look up your FA and found her web page. she has a pretty decent bio, I wrote her and I'm waiting on her reply.
And it doesn't hurt to keep building your TSLA investment. So much is coming in a very short time period. And go long on as much as you can as well. The next 5 years will be amazing.🎉
I love the idea of solving the problem of edge cases with absolute brute force of compute power. But I often wonder with neural nets which are trained on data which have a certain set of statistics - won't there always be edge cases that skirt what the neural net has seen before? That are outside it's training set. And in those cases, what does the car do? How do you ever get to L5 in that case? It just seems there will always be the corner cases that the neural net has not seen before, and misinterprets with AI. What is incorrect about my logic here? I'm looking for feedback.
I have a sense using AI generated road data for autonomous training is where action is. It’s a great way to fashion and test edge cases. I wonder how long will it take to train an FSD model with 100 exoflops of compute? Will daily updates of FSD be possible? Thanks for the video.
Gotta admit, I am sad I purchased FSD. Granted, it was only $8k at the time, but there were projections made that V3 hardware would be good enough for robotaxi, and now it is clear that it will at minimum need v4 hardware, which cannot be upgraded to from v3, or most likely v5 hardware. Still enjoy being an FSD beta tester, but that $8k could have been put to better use. I think v3 hardware may only be capable of level 2 ADAS. Maybe level 3, but we will see how the software advances and whether the v3 chips and NNs can handle the pure video, occupancy network, and probabilistic methods the are moving towards.
Better call James Douma - but great
I hope its sooner than five years!😢
The data, the compute power, but Dave, you missed another crucial element, the people. No other car OEM has the AI talents that Tesla has.
I agree that Tesla has a lead, but I think it comes more from their software and overall systems architecture. An exaflops lead is transient. Other companies will catch up, but 100 exaflops could easily be wasted if the compilers can't spread the computing load and avoid compute and data bottlenecks. Tesla has been refining this part of their technology stack.
The software has to be compatible with the existing research and industry open-source standards so Tesla can have access to advances that are happening across the entire sector. This means Tesla's tools need to be backwards compatible with a large and exponentially growing software technology stack. As a tech entrepreneur, I'm sure you understand how challenging this is.
Everybody talks about the tremendous data lead Tesla has, but that wouldn't have any benefits if there wasn't a way to refine the data and filter out the important bytes from the exabytes that are generated. Tesla is already using hundreds of exaflops of compute power because they delegate the first part if this filtering to the millions of vehicles on the road. Every car they've built can act as a peripheral processing unit because of the way Tesla's system is designed and built.
Software is king, and Tesla has the best software team that has ever existed.
I think a glacier is in the mountains, you're thinking of an iceberg. But nevertheless, great TSLA info.
Glacier? Or Iceberg
Great video and agree FSD is getting ready soon for the biggest hurdle being Liability in the USA. Fear it will make FSD crazy expensive as medical care as the liability insurance will drive cost up. Therefore my prediction is the rollout outside the USA will go much faster.
Just crazy as there is enough proof today already that the average driver, with 2 eyes only, can never beat an FSD system, alike there is no ches player anymore who can beat a computer .
How long can liability legislation, supported by many law firms just facilitate over 40k fatal casualties annually will be the nut to crack.
Enjoy your trip.
Very good explanation: the key is the computational power and the amount of data, as you say. It will be a surprise to the nay-sayers. I doubt that many saw the iPhone coming back in 2007, either.
Weird: FSD was doing well when it decided to change into a well marked unprotected left turn lane and then shot through going straight. Navigation was to go straight. Good road, overcast but very good light, clear markings. 60 km/h. It randomly screws up on the simplest things.
Many (self included) have seen this lane misbehavior. It’s not uncommon when they make big changes and remove “guardrails” to see regressions that need to be “trained out.” Looking forward to improvements!
You might be right, and my car can drive itself. However, there are so many edge cases. My car makes mistakes all the time. It gets in the wrong lane at the wrong time and it cannot get back over. I have other examples where it is doing the job but it’s so irritating to all the other drivers that it’s just not ready. I certainly don’t think it’ll be ready in 12 months. Sometimes I think it’s such a pain in the ass I don’t even want to drive with it.
The future is going to be bright thanks to Tesla!! 🙌
Was impressed by my FSD in the last drive,
first of all the map gave me a choice of routes including the late nite one I always take from the freeway to my house on surface streets, it had never given me that choice before. It might not be the most direct, but it avoids stopping at dumb red lights
Then it drove the route almost exactly how I would have driven it, and it was almost like it was copying my style of driving. (slightly agressive)
Tesla's Autonomous already is the standard, like charging
Ford and GM just won't admit it yet. Legacy has it's hands full just getting a marketable EV into production.
Ford has secured the funding and has the mind set. Jim Farley recognizes his strengths and weakness. He is customer focused .
FSD - i don't think it will take another 1-2 years. What I expect latest in 6 months an approval in min 1 country or US State, maybe limited to defined streets as Highways or Autobahn with a max Speed of e.g. 140 km/h
Dave Lee is too smart. I think what we're viewing is an AI Avatar of Dave. Just think of what can come out of a Dave Lee Avatar and FSD. It's mind boggling. I for one am all in on Lee-FSD.