Andrej Karpathy (Tesla AI lead): how Full Self Driving works, data, and Elon's super powers!
Vložit
- čas přidán 25. 07. 2024
- In it's inaugural podcast, The Robot Brains, with Pieter Abbeel (link below), interviewed Andrej Karpathy, Tesla's Artificial Intelligence lead--in other words, the guy who's in charge of getting Tesla cars to drive themselves! The wide-ranging interview touches on many topics, including Dr. Karpathy's background, his Stanford Deep Neural Network course, how he came to be at Tesla by way of Open AI, and even Elon Musk's super powers. Most fascinating was Pieter and Andrej's discussion about how Tesla's software is a paradigm shift: from hand-coded algorithms that have failed to "learn" how to drive, to 4D deep convolutional neural network architecture (software 2.0) that is put in place and trained by massive user data, the shift is nothing less than revolutionary, and according to Dr. Karpathy, Musk wants to infuse this AI tech into all facets of Tesla. This video breaks down some of the discussed matters, explaining what the two brilliant AI practitioners are discussing.
**Check out our Patreon site here: / drknowitallknows . Thanks for your support!
**Want some awesome Dr. Know-it-all merch? Check out our awesome new Merch store: drknowitall.itemorder.com/sale
For a limited time, use the code "Knows2021" to get 20% off your entire order!
**You can help support this channel with one click! We have an Amazon Affiliate link in several countries. If you click the link for your country, anything you buy from Amazon in the next several hours gives us a small commission, and costs you nothing. Thank you very much for helping support the channel by clicking our link and shopping on Amazon! (If you live in a country not listed here, please let us know and we will add it!)
* USA: amzn.to/39n5mPH
* Germany: amzn.to/2XbdxJi
* United Kingdom: amzn.to/3hGlzTR
* France: amzn.to/2KRAwXh
* Spain: amzn.to/3hJYYFV
**What do we use to shoot our videos?
-Sony alpha a7 III: amzn.to/3czV2XJ
--and lens: amzn.to/3aujOqE
-Feelworld portable field monitor: amzn.to/38yf2ah
-Neewer compact desk tripod: amzn.to/3l8yrUk
-Glidegear teleprompter: amzn.to/3rJeFkP
-Neewer dimmable LED lights: amzn.to/3qAg3oF
-Rode Wireless Go II Lavalier microphones: amzn.to/3eC9jUZ
-Rode NTK Studio Microphone: amzn.to/3vjMomW
-Focusrite Scarlette 2i2 audio interface: amzn.to/3l8vqDu
-Studio soundproofing tiles: amzn.to/3rFUtQU
-Sony MDR-7506 Professional Headphones: amzn.to/2OoDdBd
-Apple iMac 27" computer (with Final Cut Pro and Logic Pro software): amzn.to/38ybTHv
--RAM for Apple iMac: amzn.to/2PRIAJv
-DJI Mavic Mini Drone: amzn.to/2OnHCEw
-GoPro Hero 9 Black action camera: amzn.to/3vgVMrH
-Suction car mount for camera: amzn.to/3tcUfRK
**Here are a few products we've found really fun and/or useful:
* A totally fun Tesla coil toy--awesome for the whole family: amzn.to/30rvzbI
* The Humane Interface, by Jef Raskin: amzn.to/3ajM7YT
* AI Superpowers, by Kai-Fu Lee: amzn.to/3dgRSs3
* A camping mattress for your Tesla model S/3/X/Y: amzn.to/3chSRrQ
*And of course the amazing "Elon Musk: Tesla, SpaceX and the Quest for a Fantastic Future": amzn.to/36hqdmS
**If you are looking to purchase a new Tesla and would like 1,000 free Supercharger miles, just click this link and use our referral link: www.tesla.com/referral/john11286. Thank you!
**Music by Zenlee. Check out his amazing music on instagram -@zenlee_music
or CZcams - / @zenlee_music
**JOWUA:
*If you'd like a Jowua foldout mount, like the one I used to make this video, check out this link: bit.ly/3tRMSQR
If you use the code DRKNOW5, you get $5 off the price of each mount you order!
*For a more powerful wireless iPhone/Android charger for your Tesla Model 3 or Model Y, check this link: bit.ly/3b5fHSW
If you use the coupon code DRKNOW on your order, it will automatically take off the price of the 64GB MicroSD card, which is very nice!
**EVANNEX
Check out the Evannex web site: evannex.com/
If you use my discount code, KnowsEVs, you get $10 off any order over $100!
And during the month of March, 2021, Evannex is doing a March Madness sale. Check it out here: evannex.com/pages/the-march-m...
**Please ask questions in the comments, or at: DrKnowItAllKnows@gmail.com. Thank you!
Twitter: @DrKnowItAll16
Instagram: @drknowitallknows
**If you enjoy more outdoorsy, adventure type videos, please check out our sister channel, Whole Nuts and Donuts, right here: / @wholenutsanddonuts5741
Link to The Robot Brains podcast: open.spotify.com/episode/0Iuw...
Andrej Karpathy, Tesla AI Lead: how Full Self Driving works, why data is king, and what are Elon Musk's super powers!
You have a gift for interpreting highly technical news and interviews in a relatable way. More of these videos please.
My iPhone works fine fine but my iPad cant get on Tesla coul Tesla only wants FSD HAVE THE NEW program able to install on 1 cpu sounds crazy but it’s the only uthing
I can figure.
Just listened to the podcast and it is now obvious to me why Tesla is poised to be the first to solve autonomous driving!
Where can I find the podcast?
This all brings me to watching our young (1 & 4) grandchildren learning. Touch, feel, experince and remember. Great explanations.
Very fun and informative, thank you! 👍🏼
Just like listening to a great lecture on my favourite subject, well done.
Thanks for training MY wetware with your data stream
It's great to have these insights -- thank you! (I think "Software 2.0" is an industry term -- not necessarily to be confused with Tesla's FSD version 1.0 vs. 2.0 code, although it sounds like the 2.0 code uses a more Software 2.0 approach.)
I can't help thinking his name had an effect on his choice of vocation...Karpathy - Car Path y - wanting to make cars follow paths, born to do it. Podcast good, and thank you.
Car Ma?
Nice Konami code reference. You are a true man of culture.
It's hard to ignore the bug...
You should've seen me cursing that thing while editing! I ALMOST went with major post-production FX work, but... just would've taken too long. lol
Great video 🙏🏽👍🏽
Something I find interesting about full self driving is that it actually has to start off extremely flexible and as adoption increases, flexibility becomes less of a problem.
The more autonomous cars on the road, the less random human behavior the AI has to deal with.
Yup!
Lol. Yes, excluding human driving would instantly prune out many unpredictable variables.
So how big does the neural network have to be is there any way of determining the number of layers and neurons required?
Also if you are putting in a new training scenario what prevents the new training from ruining previous training?
I think DOJO is key
I don't think this V9 rewrite happens in 1.0 Software, but rather in the data architecture (so in the SW 2.0 space). But an interesting question to me is: do they have to re-train all of their NNs with this V9 change? Or is it a tweak to the trained NNs from the current 8.x version?
O Wise One,
!. Where is the "Learned memory" stored, on board or in the cloud/Dojo ?
2. If FSD is a subscription, what happens to the "Learned Memory" if one does not renew for a couple of months then subscribes for a monthly trip?
12:15 Learning distance and space first makes recognizing objects faster. The solution still has to be loads of data, but the order it learns things could be improved. General intelligence will need to learn in an efficient order.
Here's something to send to Karpathy. I watched a little girl who'd just learned to walk teaching herself about the normal force and what things are made of. Now that she was mobile it was clear she wanted to find out everything she can about her world. She toddled over to the garage palms up and pat the wall., then the door. So it's not a marshmallow, and stucco is not a form of fur. She toddled over to a bush, palms up, then the swing and slide. An FSD computer even connected to Dojo never touches anything. Maybe it should. Which is two issues. First, what can training AI about textures and materials contribute to driving, but also might there be a benefit from a non-car based source of information? London Cabbies in training ride around on bicycles with a panel for a map on the handlebars. Race car drivers walk the course before driving it.
Something else.The Cybertruck off road. The sand on those roads is almost always made out of the boulders on the road and the cliff faces. So the contrast can sometimes be extremely low. I'm hoping that the Cybertruck can do some boulder climbing, and be able to do it FSD. I'm not sure the visual system will be able to handle the low contrast between the elements in front of, and under the vehicle. Might be a lidar situation. (which of course doesn't mean Tesla is wrong about road driving and lidar. I've driven at least a million miles without it.)
Great job!!!! I think Tesla's approach to FSD is better than the other alternatives. But I see some problems:
1) Perhaps the problem of the long tail has no solution. The solution is not fixed, the rules, the customs, and conventions evolve. By the time the neural network parameters are optimized, the solution could have moved.
2) From the interview, I understood that data acquisition is not completely automated. It must be curated by humans. I'm not sure this part is scalable.
3) What happened when you need to intervene once a month? Drivers will still need to give their full attention. Could they?
I would love to know your opinion
Dr. Know, the AI research model is dependent on binary computer systems. It's a limiting factor for AI in data, processing and self-learning behavior. Next time ask Karpathy how quantum computing might change AI research and in particular FSD. I'll be interested in hearing what his answer is.
Wasn't jayprakash Vijayan the "architect" of Tesla's autopilot
what different ways are there to train neural netwroks? It would make an interesting episode
A good idea from you.
How long is your attention span?
@@silberlinie I want to know too
@@amantedecapaldi4037 Well, a nice channel with one of the great experts who can explain pleasantly is here: czcams.com/channels/ZHmQk67mSJgfCCTn7xBfew.html
If humans become at least partially cybernetic, using Elon's neuralink for example, their enhanced abilities can be added to those of nonhuman machines. Eight billion additions is quite an improvement. Now, when quantum computing is more prevalent, well that's very dramatic. It will revolutionize the field if AI. Quantum computing is the ultimate linchpin for this century. I often say that the way we lived in the year 2000 has absolutely nothing in common with the year 2100. Not one thing. But a good deal of that is by 2050. Now think about transportation if we have a fair number of quantum computers available like personal devices or embedded. Perhaps you will end up within far sight of the reality of 2050.
@@PlanetEarth3141 It is eight billion Multiply-Accumulate (MAC)
Has the Robot Brains link been removed from description?
why do you put the link to the source in the bottom of the desciption after 100 affiliate links ...
What is that giant bug on the wall and ceiling?? Starting just before the 7:00 mark
Stink bug. And yes, they really stink if you squish them.
Stink bug... they are called “bärfis” in Swedish. Translated it would be “berry fart”
@@DrKnowitallKnows I recently bought a Dyson V11 and it is SO good for killing stink bugs. The high suction causes it to die on contact from being slammed into the vacuum wall when it’s sucked up the tube. And since it’s completely HEPA sealed there’s zero stink. Sucked up 8 of them the other day in a matter of minutes and no smell at all - truly amazing if you live somewhere where there’s a lot of stink bugs (we get tons here in Massachusetts in the warmer months).
@@Brendon471 I need to try that on bedbugs.
Elon Musk + AI = innovation at all levels and scenarios eg wiper, air conditioning, 4680 battery, factory design etc
Maybe talk to Andrej about your ideas for teaching AI?
You may know that the different Waymo/Tesla approaches between learning from pre-programmed mapping as opposed to massive input of data goes back to the difference between Plato, who thought that understanding and recognition were only possible if we assume that humans, prior to observation, have inborn patterns to which we compare perceptions; whereas Aristotle thought that patterns were built from the ground up by lots of observations--Waymo and Tesla, respectively. It may sound overly pedantic but when I gingerly brought the relationship between ancient and modern up to a professor of AI who works for Waymo, and expected a brush-off, she immediately said, "exactly." It is fascinating to me that the experiments in neural networks has brought us full circle back to the insights of these ancient thinkers. The Plato/Aristotle dispute continued through the history of western philosophy in the debate between rationalists and empiricists, respectively. BTW, I'm not a weirdo that believes in the mysterious knowledge of ancient civilizations; but I will confess to teaching philosophy for 30 years. I just think those were two smart dudes.
I am very interested by your analogy to nuns slapping wrists in AI. The thing is, that whether you program the computer in advance, like Waymo, or start letting it figure things out by itself, like Tesla, and even if you are able to achieve "vacation mode," there is still a human doing the ultimate judging of success. And we still don't know how the human mind does this.
Regarding your final paragraph - in a semi general AI future, will we always want experts/the media/researchers be the final judges regarding the success/bias/unbiasedness of an AI implementation? What will happen when an AI makes a "Move 37" (see game two in AlphaGo Vs the World) but in an implementation field that is not as clear for humans to objectively evaluate? Let's say possible candidates for round two in a job application process? The neural net perhaps chooses some candidate that a HR specialist sees as a clear mistake, lacking a specific set of official competences, but were the neural net has identified some underlying correlates between that specific skillset and some other achievements that the candidate has shown. These correlates could be outside the realm of human prior knowledge/possible understanding and the expert giving the final judgement risks to not utilise the full potential of the AI-powered tool?
These sort of EXPERT vs "Move 37" conundrums will start popping up all over the place. For example, the FSD Betas current behaviour of "California roll" at stop signs: we as humans would consider this as sloppy driver behaviour and leading to unnecessary risk and thereby something that we need to train away (here comes the Nun with the ruler) from future versions. But it could be that the array of sensors, with super human fps identifications of possible cars/threats, does not need this extreme deceleration. Instead the FSD has identified, either from shadow mode or from DOJO simulation that it is actually better/safer to drive at 2-6 mph at intersections. Should we as drivers/machine learning "experts" a priori say that this is "bad behaviour" since it is not as humanlike as we want?
@@erik.dahlberg This is a very interesting issue and quite beyond what I was considering (although, remarkably, not beyond what Plato and Aristotle were discussing). I was merely speaking of the nuts and bolts of commonly understood reality that a human would, as in the fire hydrant example, have no problem with. The idea that two or three examples are sufficient for a human but millions may be required for AI is quite mysterious. What is the brain instantly gathering from the surroundings or the appearance of a never before seen object that are invisible in incomprehensible to the AI? Your example of the Californian stop is an excellent one when extrapolated, as you do, to bigger questions like, say, the trolley problem in ethics, or, say, government decisions.
@@terrulian Indeed, those Greeks sure had foresight, and the inclination for deep thought. But why is there that pigeons to us seem to be feathered fools, almost governed by a "drunken walk"algoritms and ravens instead seem to be cunning "tool makers" or even pranksters, given that they are both birds of similar sizes and function, the sizes of their brains are fairly similar, they are both adapted/designed by evolution and have overall mastered the same capabilities: to find food, to find partners, to take care of their children, to fly safely, to effectively identify and avoid threats, while minimising type 1/type2-errors. The latter has probably some type of internal sense of worth/capability for empathy of its peers, the prior only walks for the closest bread crump. Which of them excels in performing the overall charade of being a bird? Hard to tell. The raven is probably bored more often though.
Chimpanzees would by any reasonable human standard be considered to mostly be raging psychopaths in disguise - that's why we keep them in cages when near civilisation - but they seem to have super human working memory, shaming even the most Adderall boosted IvyLeague student, for reasons unknown to us? To more effectively sort nuts by category perhaps? We are walking in the blind here, even when properly understanding our closest cousins.
Therefore it seems to be a really hard problem to a priori set up criteria how to aim for any level of complexity of a neural net for any specific problem to solve, or behaviour to mimic/surpass.
So, how do we build the foundations for robust AI systems, that both perform well in terms of naive human evaluation standards and are explainable and/or understandable by the engineers that "built" them, when we can't even tier out wether or not a higher minimum wage is good or not for the society/GDP/employment? I think we will see a future were we will have to come to terms with the idea that most new, modern technology and policy making trade offs will seem more and more like magic again to most members of society, even those very experts that are involved in any particular change/implementation. Instead, we can reprioritise that newfound time and energy to spend on human connection and interaction. The alternative will be to instead ignore superior solutions that we can't properly understand and therefore knowingly endure more suffering, like energy grid brown outs and traffic deaths. It is both a bleak and amazing prospect for the future.
Hey so Elon has said LIDAR was a crutch. Now that Tesla is moving to vision only, Do you think that radar was/is a crutch too?
I don't think so. Radar can "see" through rain/snow/fog.
Something like that is definitely necessary to get to super-human driving skills.
I think they want to remove the radar for now from training. Maybe its preventing the neural net from truly getting the depth from only vision. Just a guess.
The end goal could also be that the neural net decides itself when and how to use the radar.
Every time I watch that intro... I see a minion's butt in the pear.. and the glasses make it look like he is in a bikini XD
How would you harmonize that humans have different personalities, feelings, likes and dislikes and because of that you need different teaching techniques to accomplish your task. Computers have none of these qualities so why would you need different techniques? I understand that different systems have different architectures, are not the technique the same? Does not the decision trees look the same just different coding depending on language? Please continue to have such informative discussions. Thank you for your hard work.
lit
I want to see how fsd handles 4 way intersections with NO stop signs. Yes, NO stop signs at all. You have to first know of these intersections. Then treat them like there a yield sign. I made sure when I taught my daughter's how to drive, I taught them to looks and see what signs the cross traffic has. The ai needs to do the same. I'm due there are very few of these intersections, but this is an edge case
I have never seen a 4 way intersection with no stop signs(or failed stoplights)...where is this? I want to see it in google maps
@@smoot96 Happy to help - check out the intersection of Lewis Ave and E 12th St in La Junta CO 81050 on Google Maps. There are many examples in this neighborhood.
I was referring to the USA. I know All bets are off outside the US. However, I'm not too concerned about those situations. I think that's just a software issue where they can determine the projected path of oncoming vehicles based on speed/time, then act accordingly.
My biggest concern is how it will handle situations where the cameras cannot see due to obstructions such as bus stop shelters, bushes, trees and even walls.
@@smoot96 Here you go. I have seen other in other towns. goo.gl/maps/rFhcXh8d1GrhgGMM8
@9:40 *giggle* you disagree? hoo boy!!
pleasing to my ear? or closer to reality?
Nice spider walking on the background
How hilarious would it be if Elon is able to use Dojo to create the first production fusion reactor.
We are not born with memory but we are born with specific evolutionary algorithms honed to organise such memory in a effective manor.
Excellent as always.
The human mind and a neural network may have some similarities as you say, but there is one tremendous difference: The human mind operates a human body during every conscious moment, and that human body lives and moves in the ever-changing world. A computer knows nothing about the world and has to be taught even the most obvious things, like that objects are subject to gravity. We can code the law of gravity, but how do you tell a computer which way is down? There is no easy way to explain this so you need a sensor, as plants and animals already possess. This is true of every other sense as well, and some of these senses are not well understood even in humans and animals. Self-driving is complicated to analyze because the world is complicated; but the car has no sense of smell, taste, or, so far as I know, sound. Like the neural network's inputs from millions of driven miles, the human body has much more complicated inputs from five or six senses and these are active every waking moment. The car has essentially, so far as I know, sight, a g-meter, and an accelerometer. We rave about the amount of data Tesla is collecting, but this is levels of magnitude simpler than the amount of data collected by the human brain every instant. Physically navigating the universe is only a minor activity of the human brain, one that we spend little conscious energy on, and we struggle just to accomplish even that with self-driving.
Another mode that as far as I know, we don't know how to do, is telling a computer what its goal should be beyond very simplistic, mechanical objectives. Still restricting ourselves to driving, we can tell a computer we want to take the fastest or safest route. But how would we tell it how to take the most pleasant route? Obviously, this is one of the elements of driving that many of us enjoy. How would you program it?
Please understand I'm a big fan of the fantastic new developments in AI, and look forward to more amazing progress. The brain has a head start of a few hundred million years. Things are progressing quickly, but it may take awhile.
Absolutely. Might've been a bit cavalier with that statement, but I'm very much of the opinion our brains are very, very complex neural networks, and fundamentally not that different from the mathematical ones we've made. Of course we have immensely more neurons than any machine, which is our advantage. As for experiencing the world, I'm with you there: no computer will ever have a chance of anything like intelligence until/unless it directly interacts with the world. Robots in other words. But Teslas are robots: they have sensors and also actuators (steering, acceleration, braking), so there is at least a rudimentary interaction with the world. As you say, it'll take a while for computers to catch up to millions of years head start in our heads!
Do we know if Tesla or anyone else is working on voice recognition routing? As in "take me to Santa Cruz by way of the Great Highway" (which one friend calls a "good" highway, not really "great.") Presumably the route would be displayed and could be modified.
Really great points. In terms of simplistic goals with seemingly advanced behaviour, I don't think it matters if a future FSD neural net has really reductive input data related to a few type of sensors and binary evaluation (good/bad) - if it behaves as if it has a proper understanding of the world, drives safer than 99+ of all human drivers, and does this in a way that we as passengers can accept (while enjoying our productive/non productive activities without nausea or fear for our lives). If the AI will mimic human behaviour good enough, that is enough.
BUT, if you are talking about other future applications, let's say an AI teacher, psychiatrist or assistant living nurse, that is a completely different ball game. Simulated empathy with extremely reductive inputs is super scary, since we as humans are so easily fooled. This has been the case since the Elisa chatbot days of the 60s. To reduce the authenticity of human emotions/suffering/joy/love/empathy to an x array vector and start running the linear algebra sucks as vision for the future.
@@erik.dahlberg Like an automotive Turing test. There is some You Tube video where a woman is following her husband in his Tesla, trying to guess during each segment whether he was using FSD or driving himself. She did better than 50% but wasn't perfect.
I think they're using AI to drive their next-level manufacturing. They did some magic fluid analytics to get the molten aluminum into the giant castings with close to 0 defect rate. Pretty sure it helped them with the design of the octo valve.
Tesla’s dojo will complete with Palantir
Hi
wow I am so early
I strongly want a comment from dr Know it all on this comment:
My fear on dark and foggy forest roads in Norway is a moose suddenly jumping into the road. I cannot see it until its close to me. Will it really be enough with vision to detect this danger? Another thing I am worried about, is the winter roads when its about -10° C and the air temperature suddenly goes to +5° and it starts raining. The rain freezes imediately on the ground and the car becomes a toboggan without steering. I would suggest Tesla to have an infrared beam sensor to tell the AI what's going on. I hope to have comments to this...
Keep in mind all Tesla’s have radar which can see up to 525 feet (160m) ahead through rain, snow or fog. It would detect the moose as a large object ahead and the car would preemptively slow down until the cameras can see the object and safely drive around it. So that shouldn’t be much of an issue.
As for winter roads, Yandex has already been driving their self driving cars on icy and snow covered roads safely, so they’ve proven you can train a neural net to handle winter roads safely (slowing down when temperatures drop and road conditions are wet, and use brakes in a different manner so as not to lose control). Search Yandex on CZcams and watch their recent video showing their latest progress. It should ease your concern for self driving in conditions which become icy.
@@Brendon471 I may have misunderstood - I thought Elon just said that Tesla would just use "pure vision FSD". www.teslarati.com/tesla-pure-vision-fsd-patent-elon-musk/
Eh if Tesla wanted cars to look like future cars, they probably should not have hired the Mazda designer. They really all look like Mazda's, and is a huge deducter to the value proposition.
Our congress has unlimited resources to always come to the slowest and worst decision.
Oh, so true and they can be depended on to do their worst.
advise skipping first eight minutes
@4:05 Not for Bernie Sanders...
Please try not to script your content... I end up distracted watching you read rather than listening to what you are saying. It’s much more engaging when you are speaking freely
@9:35 So if you don't learn your kid vision or how to remember stuff they will never be able to see or remember what they see / saw? :D
This is definetly built in our dna and not something we "learn" over time. It's just that we are not done when we pop out and the brain needs to develop more (and it does so with or without learning).
spider alert
Run this at .75 if you feel he’s pouring the info too fast into your head.
Govt's to regulate congested/polluted cities with carbon neutral laws. London law for congestion/carbon ULEZ, Tesla is positioned/ all countries with dense pop cities, Starlink w/ G5 allow networks to track everything in realtime. Neurlink/allows disabled to travel/ All travelers will buy FSD/ NET/TESLA.
It was, in fact, not clear to the layman.
those are nice glasses
You have way too much junk under your heading for each video. Why is Spotify podcast last vs first? Bad design 🤔😂
9:39 YOU ARE WRONG!!! (for once lol). It's well known that babies have innate facial recognition, for example. We have "cheat codes" that allow us to learn very quickly, about a VERY SPECIFIC set of kinda predefined things.
1984 is a wonderful movie/book...the 1% tell us what to think they tell us 2+2=5.......nop.... no thanks ill use my own brain
a lot of the video, is repeating the same stuff over n over, without analytical insight into the neural network process, and how that is transferred to the car. many many words, few thoughts.
I disagree. I think we are born with it. Many animals can walk or run shortly after birth.
When you are doing your video you are conscious that it is no good and why should people listen to you. Tell that inner voice that it is a liar and tell yourself that people really want to listen to you..... because we are really interested. Your inner voice is LYING!!!!! kkkk
There's just one problem - Tesla's FSD does *not* work. It's a glorified cruise control that is not capable of true autonomous self-driving.
You are right, the name has been misleading for a long time, but finally it is pretty close with the limited beta version.
@@casperhansen826 Tesla is not even close to true FSD. Musk said they'd show a demo drive from one coast to the other. It's 3 years late and counting. FSD is more than keeping the car going somewhat straight and following lanes.
@@user-xv4id9xx7u So you haven't seen the limited beta version on youtube, you should, it is quite impressive.
FSD with ai frees a person from driving so a person can do something else such as playing video game, but driving is a pleasure as we can see the beauty of God's creation of the world and think how we take God for granted
Non sequitur
I have been a great fan of Musk for years, but I fear his vision of the future is in danger of being taken over by toys and not practicality.
Cars are transport. It’s all very well having video games in them and being able to make them fart but when basic functions like windscreen wipers and autoparking just don’t work properly things are not going well.
Just because it’s possible to do things via a touchscreen - even when they DO work - doesn’t make it better than a lever or knob when in a killing machine.