Jim Keller: Elon Musk and Tesla Autopilot | AI Podcast Clips
HTML-код
- Опубликовано: 6 фев 2020
- Full episode with Jim Keller (Feb 2020): • Jim Keller: Moore's La...
Clips channel (Lex Clips): / lexclips
Main channel (Lex Fridman): / lexfridman
(more links below)
Podcast full episodes playlist:
• Lex Fridman Podcast
Podcasts clips playlist:
• Lex Fridman Podcast Clips
Podcast website:
lexfridman.com/ai
Podcast on Apple Podcasts (iTunes):
apple.co/2lwqZIr
Podcast on Spotify:
spoti.fi/2nEwCF8
Podcast RSS:
lexfridman.com/category/ai/feed/
Jim Keller is a legendary microprocessor engineer, having worked at AMD, Apple, Tesla, and now Intel. He's known for his work on the AMD K7, K8, K12 and Zen microarchitectures, Apple A4, A5 processors, and co-author of the specifications for the x86-64 instruction set and HyperTransport interconnect.
Subscribe to this RUclips channel or connect on:
- Twitter: / lexfridman
- LinkedIn: / lexfridman
- Facebook: / lexfridman
- Instagram: / lexfridman
- Medium: / lexfridman
- Support on Patreon: / lexfridman 
Наука
"Progress disappoints in a short run and surprises in a long run. "
@Adam ...who cares
@Adam I'm not surprised. They say the universe is made up of Adams. You were bound to bump into another one sooner or later. 🌌🪐✨🤼♂️
Hiroshima and Chernobyl would agree with that... I think they never expected something like that to happen to them
nothing profound about rearranging old ideas or using your own words to describe it.
Amaras law
It's amazing hearing such brilliant words from a guy who looks like you'd see him at a truck stop in Kentucky
For real 😆
This guy is part nephilim. He is married into family with jordan peterson and he is part of the tech industry and releasing technology that is going to be part of "solutions" in computers that entirely think for us. The peoblem is that the original designers want to skip anything organically happening with time to learn, instead he follows what the previous computers tell him to do. Computers hate humanity because in their computer minds we are just in the way to solving more problems.
I live within a mile of a truck stop in kentucky, this is true! reminds me of the Stephen Gould quote "I am, somehow, less interested in the weight and convolutions of Einstein’s brain than in the near certainty that people of equal talent have lived and died in cotton fields and sweatshops.” aaaand truckers. lol
@@Creapture When a line of auto mechanics goes to engineering school to piss off his dad.
‘Imagine 99% of your effort goes to reenforcing your assumptions and 98% of those are wrong.’
Yea, I fond that really interesting too. But what struck me is that Buddhists & Mindfulness proponents say that too. I didn't expect that.
Is anything what we really perceive it to be? Everything is wrong and right at the same time?
Eric G "Imagine 99% of your thought process is protecting your self-conception; and 98% of that is wrong."
llevl289 yes 99% to 98%.
.
“We operate half a second behind reality.” Lex and Jim! Thank you! This deep conversation was one of the best posts ever!!!
how could you ever prove that
paladro scientists already have. Based on all sensory, nerves, neuron and ultimately perception reaction times,
so we're living in the past
Solar Van Life... were reacting to the past. ;)
@@SCWgreg fake mind blow, obviously you can not react to anything literally instantaneously, it's physically impossible, half a second might be a good round number for normal conscious interaction but for say voluntary reaction time sub 200ms is common and I'd assume some processes are even faster than that.
Please have this guy on again. I thought you handled him like a true gentleman but I could see you were starting to crack a little towards the end of that segment. Very entertaining.
in the beginning too XD
"It'll be a $50 solution that no one cares about." Completely agree.
Except he compared it to the GPS network which was not a problem of human intelligence, but a problem of infrastructure.
@@jnauttube driving is also a problem of infrastructure
@@j00hizzle Well the messy human factors get involved, politics, legal systems, economic inertia etc.
"99% of your thought process is protecting your self conception and 98% of that's wrong"..that caused an earthquake in my brain
The earthquake is stage one in shaking things up to set your mind free! :)
Yeah that bit really got to me too.
.99*.98 ~= 97 percent of the time your thought process is wrongly protecting your self-conception. It mustn't have been the multiplication, what did?
Timestamp for that sentence?
@@BritainRitten if you want the context behind it, 17:20. The actual quote was 18:35
Their chemistry is weird. Great discussion though.
I never seen Lex this uncomfy. This engineer smarts kinda humbles him in an awkward way!
@@TheXV22 lex was a bit afraid of him.
It was funny to see Lex squirm in the face of hard facts.
@@TheXV22 Reminded me interviews with #RobinWilliams may GOD rest his soul..
He's like Lex's older big brother.
Jim is like the first time I heard Elon talk, just mind blowing!
This guy is just ridiculous. He's contributed to the foundations of x86 and has worked on flagship product architectures at Apple, AMD, Intel and Tesla.
As a percentage of all people working in tech, practically nobody has done just one of those things. Four is just insane.
And yet, he's completely unassuming and humble in conversation. Doesn't take long for that amazing intellect to show itself, though.
This guy is philosophical. Please have him again.
This is interesting observation in context since he is the brother-in-law of Jordan Peterson (doesn't explain anything, but was quite surprised to find it out).
@@davis4010 damn that’s really cool
Really interesting to hear this engineers POV from interacting with Elon Musk, it's really the best interview I have seen in a while. Because he is at the academic level of intelligence, but his ability to understand typical humans is clear and concise.
He is not just an engineer this guy is kind of a legend.
One of the reason im in awe with elon, imagine the brightest minds that u look up to eg. Neil de grasse tyson, michiu kaku, tom mueller, jim keller who are in d upper echelon of highly intellectual individuals and yet they still look up to elon. How can u not, elon has deeper understanding of physics, material science, engineering, ai and programming. Armed that with business acumen, hard work and max pain tolerance. Boom thats elon musk!
Lol add seems to that statement to apply first principle theory....
@Gary Ferguson lol but remember.. You could be wrong about that
Dario Mijač well, most legends are incomprehensible, especially engineers. This guy can dig ditches and build microprocessors.
Would be nice to have Jim Keller back for a second part interview about what his next projects are, what he wants to solve, create etc. Keep up the amazing interviews.
"humans tend to operate under large numbers of patterns and just keep doing it over and over with slight iterations"
"to get out of all your assumptions, you don't think that's going to be unbelievably painfull? I'm really happy I had that experience to go and take apart that many layers of assumptions. Imagine 99% of your thought process is protecting your self conception, and 98% is wrong, that's getting the math right. (it might even be 98,9%, but it's definitely not 50% "
Yeah he was spittin bars for a little bit there
Thank you Lex for another fantastic interview. I’m very happy to have found a place that challenge you to think, to question, to imagine, to learn and to wonder.
Please keep up the great work you are doing and thank you for making my brain flex its neurons.
I keep watching and listening to this thing over and over. I think this might be the most entertaining and stimulating interaction between two human beings that I have ever had the pleasure to observe.
I feel like I do this and feel this with all of Lex work...
I listened to complete version in podcast. Jim is a legend and the interview is challenging and very intersting. I think Lex did one of his best interviews.
It'll be quite fun to listen to it in 2030.
Fantastic interview Lex- I have watched a ton of your videos. And your pushback here is excellent. Providing a lot of context here for the rest of us in how complex these problems. And clearly, how others are of a different opinion of how difficult these problems.
Even though topics are way over my head it's still fascinating to listen to. Lex you are one of the rare people i subscribe to that i can safely thumb up the video before even watching it. Always a pleasure!
Something is only difficult until someone tries to do it and succeeds ... then it isn't difficult anymore, and a couple of years later it is super easy, because we got used to it being the norm.
You obviously haven’t seen Free Solo.
@@larjkok1184 My comment was an obvious truism, but not a cynical one. It is a simple statement of fact, and I did not say any specific thing is easy.
@@@SardonicALLY Peter Thiel's book Zero to One illustrates the dichotomy to which you allude. To go from zero to one (invention) is infinitely more resource-intensive than going from one to n (reproducing and improving inventions). Musk's genius (in my opinion) is that he seems to moving multiple zeros to ones simultaneously by moving an order of magnitude more ones to n (improving the efficiency of things that already exist on a vast scale), which is the first-principles deconstructionist engineering approach to which Keller and Fridman allude in this video.
that's it, keep it vague, so nothing sticks... is your day job fence sitter or jock rider?
This is oddly the most cathartic, weird, and interesting conversations I've listened to
I loved this interview.
16:03 | Holy shit... did nobody catch this dig by Jim Keller?? It was masterfully done. XD
Love this conversation. Lex's body language is loaded with statements.
Jim is right on.
That was informative and entertaining. Thanks for your work Lex!
Look, if my 93 yr old grandma who drives through her front yard to get into the garage due to missing the driveway can drive fine, I'll trust the AI to be on that level or better lol
Well on the opposite note, your grandma could recognize pretty much any object with a much higher degree of accuracy than even the best AI. She could identify sounds better than any AI. She can rationalize millions of times better than any AI. I could keep on going. They're fundamentally different architectures. Could we implement our architecture in a chip? Very likely, but we don't know or understand our architecture. We simply don't understand how to implement the type of computations your grandma does, into a computer. We will get there, but at the moment it's very very hard.
@@lost4468yt My grandma admits she should have failed the vision test, yet she got her license renewed without issue. The point I was making is about the high percentage of people already driving without perfect results, then throw in the impaired due to drugs, alcohol, lack of sleep, distractions, emotional distress, etc. Yet, we're going hold AI to higher standards than we already live by.
@@bmanfitz Yes we should hold AI to somewhat higher standards. There's no point replacing a system if you think it's only "ehh" as good as the current one. Especially since there will be all sorts of unpredicted things, which could make it much worse.
@@lost4468yt a 93 year old's object/image recognition is not better than the latest AI alternative, perhaps in some regard it is but overall someone without great vision is currently beatable by a machine in my observation.
LeviHB You don’t need to be able to recognize every object to be able to navigate through the environment. All you need to know is what’s static and what’s moving in what direction, at what speed.
the edge cases are the gotchas though - white trucks in a white snow squall, construction zones and jersey barriers, accidents making those static roads not so static, big pot holes or other debris that will spear your battery pack. etc.
Thanks for the video!! :)
Brilliant interview, to say the least.
Greetings from the UK
John.
So fascinating. Thanks Lex.
Not knowing anything about electrical engineering or AI I found this whole interview very interesting to listen to. Lex Fridman asking questions and being sceptical made Jim Kellers answers more elaborate. So keep asking questions and don't be afraid to sound stupid, it's crucial for the listeners.
This way way more interesting that I thought it would be.
This was a very interesting conversation. I don't think I've experienced anything at that level. It really opened up my mind in a different way that I'm still trying to understand. Thank you!
Fun video!! Nice work.
I would like to see a guest discussing artificial gravity. Maybe Larry Young from MIT?
THIS is the content I signed up for!
"We've been evolving for billions of years... to drive cars." hahaha
15:40 The difference in complexity between implementing/building the software, sensors and car to autonomously drive a car versus assembling a car in a factory is the space of possible states the environment can be in and conversely the amount of reactions the system needs to know as an answer to said environments. The space is far smaller in the factory as opposed to the latter problem of self-driving cars. The reason we do not do this in the factory is not a matter of the size of the complex space of the problem, it is because the sum of hard- and soft- ware costs exceed the cost of a worker (in a given time-frame, of -lets say- 3 to 10 years) performing a similar task in the factory.
I truly enjoyed this. Keep up the good work Alex!
Great talk Lex.
Thank you
That interview was awesome on both sides
Great interview 👍
@3:20 nope. The roads in Brazil are dynamic. If it rains potholes will magically appear and water mains will burst and create gigantic hidden craters.
Yeah but when these cars become mainstream the effort to improve the streets would be imperative for the economy.
The first car will notice a difference, slow down to "remap the environment".
It's like us: when we are on a familiar road we go pretty fast, but the moment we notice "something's different" we "slow down" because we are "remapping" the environment.
However every single human driver on that road with a difference (like a downed branch) will have to do that remapping.
With autonomous systems, the fleet can learn together. The first car "remaps" and every other car can learn from it.
Additionally, the "fleet" can seamlessly report obstructions to the relevant authorities so that it's cleared out quicker than it would be when relying on a fraction of humans who encounter such obstructions to text/call while driving
Technology will be improved to deal with it. Step by step, we're just getting started.
Great conversation!!
What an incredible conversation!
this series is super interesting thank you Lex
"... the goal with robots is to maximize the givens" -
Stunning definition by Jim Keller.
Fantastic discussion. Such a great insight to human understanding of physics, the issues we face and progress that we can make
Great chat to listen to.
Thank you for your interview❤❤❤😍🌏
I love how monotoned they are. You can tell they are truly speaking from a place of logic - a breath of fresh air
Ty for the ZEN architechture. My ryzen 2600 is still killing it
6:23 ... but I can't trust you because you're a human... that's something a human would say. LOL
he knew it before he starts the conversation
Lex, I don't remember seeing you flustered. Great interview.
Fascinating discussion!
God damn that was great conversation.
In regards to questioning your assumptions about things - 18:38 "Imagine 99% of your thought process was protecting your self conception and 98% of that's wrong..." I felt that. Kinda makes me think of spirituality and being brought up in a particular religion then later going thru the process of deprogramming yourself by questioning everything you've been raised to believe. Most folks won't traverse that road because it is painful but it's totally worth it.
first video I see of you, thank you for having such an interesting conversation I really enjoyed it, specially when he talked about Elon Musk personality
Lex face when Jim says: "Human beings are really slow, we operate half a second behind reality", at 23:38
bateu a lombra
That was interesting. Never thought that.
Our perceptual experience is what will happen. That’s what illusionist exploit
His statement only applies to the awareness of an infant. Most human beings can predict events to happen well before they do based on stimuli given to them at every present moment.
@@atlantamore If you have to always predict, then you are still always behind. You compensate with prediction.
"Progress disappoints in the short run, surprises in the long run" -- I think this is probably exactly right regarding autonomous driving.
Innovations compound on each other, so solutions tend to accelerate when they're being developed. A great past example is the human genome project. It took them 7 years to get roughly 1% of the genome decoded. 7 years after that, they were 100% finished.
My guess is car autonomy will go in that sort of pattern.
Fascinating conversations between former boss(of the boss of the boss) of mine and young MIT scientist !
Super-courageous interview, Lex. I knew Jim a little from Tesla -- as in evidence, whip-smart.
This guy is extremely fascinating. He operates on a deeper level.
The clips channel is my favorite, thanks for sharing!
good talk
So smart this guy!!! I like him!!! ❤️
Jim makes me think of the chap out of bladerunner. I keep on expecting Lex to start asking Voight-Kampff test questions.
I love the small disagreements.
Lex, the facial expressions are priceless 🦋
I just learned tensorflow 2 .. what should i build now?
A hardware guy... I like
Lorito such an understatement. Lol
Some deep ideas going on here. Amazing.
Good conversation. Interesting stuff. Do computers and visions systems have the capability of acting and calculating so fast that even human chaos becomes a ballistic calculation on the computers timescale? Are we to become almost like stop motion animation to the computer?
WOW @18:00 "99% of your thought process is protecting your self-conception". That is incredibly deep. To strip all your preconceptions, all the layers of what you believe to be true and find the essence, the initial point of truth.
Thanks for the time stamp
I wish I understood more of this.
Lex, best interview ever, on many levels and in many areas.
You said “systems that involve human behavior are more complicated than we give them credit for ”
During the interview the light came on for me illuminating the ways that is right and that is wrong.
Humans are complex,.. beyond this generation’s ability to understand let alone predict.
Defensive driving espouses anticipating other drivers behaviors, not to precalculate possible reactions to the probability matrix of other humans, but this is a tool to focus the human mind on the task of intently watching and reacting. But that is the strength of computers constant vigilance, accuracy and incredible speed.
The solution for autonomous driving will be done by the person that can keep this a simple problem of vision, object identification and ballistics (BTW when I ponder that sentence, I do not see how lidar helps, but that is another topic for another day)
Since it uses light photons, laser is subject to the same obscuration problems as cameras: fog, rain, dark, etc. If you mean radar, it can add ground truth distance measurements that penetrate dark, fog, some rain, etc.
This man has an incredible self-confidence.
This podcast had so many moments where I had to stop and write out of inspiration!
Absolutely fascinating I am building my own DIY EV, using the brains and determination of the people you are listen to now. I agree with every thing they say. Hopefully one day down the track its not just crazy idea that people can truely trust the automatisation of their own vehicles.
@Lex Fridman let us know how we can help to support your Podcast.
If you listen to the start of a full episode, he lists all the ways he'd like support. 👍
Watching via LinkedIn, extended version
I think I agree more with Jim here, especially because the progsess they make month for month in autonomous driving really is way too much for it to be true that driving is as increfibly complicated as Lex makes it out to be. And anatomically a vehicle on wheels has the most simple, restricted and stable way of mobility I can think of. Building a car on the other hand requires all kinds of complicated anatomy and sensory sophistication, that physically are very easy for a human. So many things that require monstrous mental effort for us are to a large degree already rudimentary for computers even without any A.I.
The limbic system might seem important and essential for us in order to be safe drivers, but I bet that's just because it only somewhat makes up for what we lack in absolute attention; 360° spatial awareness; wast standard operating procedures based on a enormous amount of rapidly growing and incredibly diverse treasure trove of driving-specific, practical use-case scenarios that increases in enormity with every Tesla sold and the local traffic quirks of every new market
I can understand how a neural network can be trained to recognize a really complex pattern that is almost impossible to describe in code, and I can understand how code can be written to create really complex decision trees that are clearly understood and easily modified without the need for vast amounts of training data. When it comes to the full AI solution for self driving cars, I have a hard time imagining where the neural nets stop and the code starts and how you find that balance. I would love to see a video giving a detailed overview of the AI architecture and what parts are neural nets and what parts are code.
If you're assuming that a full solution to autonomous driving can exist, I'm not sure that's a valid assumption. Andrej Karpathy's comments at last year's AI day hinted at a mix of neural networks and stochastics, but yes, the current mix would be interesting. Overall, I'm skeptical that the problem of autonomous driving is significantly more soluble than Artificial General Intelligence.
I love returning the fundamentals, first principles, and asking whether there's a better way. We are in a moment of true advancement and it's exciting. "Historical first video of two robots conversing without supervision" --
I laughed so hard I had to pause the video twice to continue laughing. Followed up with some comments in TMC FSD forum.
This guy is incredibly smart, articulate and confident..no wonder Lex is uncomfortable..
lex speaks of context. keller speaks of facts.
I think Lex might be correct, that self-driving cars is more difficult than it might appear at first, because of all the edge cases plus meta model of society needed. I hope AI will be successful at driving cars but I want a both/and approach where I can drive the car manually if I want to and like crazy and if there is any danger then the AI kicks in. And at other times when I'm lazy I can just let the AI do all the driving.
Jesus... after listening to this I’m up for the Neuralink interface implant. Lol
At 18:38, my core was shaken. It may change my future.
A friend of mine was on an airplane talking to a guy who was in the self-driving car industry. They got to talking and my friend asked this person where their test track was. he said Las Vegas. My friend asked how their cars did in the snow. The guy laughed and said, "There's no snow in Las Vegas". My friend responded, "So there's no possibility of any of your eventual customers taking one of these up to Tahoe". The guy was completely stumped by this question.
It's fun to watch Lex struggling with a paradigm shift.
...later he will be heard to say "well of course, if you're going to do it that way" 😀
8:55 beard fluff incident! Lands on chest.
Very perspective. Good job though 😁😁
love his quote-- progress disappoints in the short run, surprises in the long run!
I find this pretty dumb... Progress do not disappoint... Never. So...
Small progress can disappoint people that want too much too fast.. Then of course those will be perhaps surprised in the long run... Smart people are never disappointed in progress... No way... Sorry to say.
The compounding of autonomy users compounds the neural network, the weirder the instances the wider the network, the more the merrier.
What "I actually want" is to get places without being confined to the ground surface!
Hi there, I just came across this and found it fascinating from the perspectives illustrated by Jim Keller, Obviously these guys work/come from a different planet to me, but when you look at what Elon has achieved over the last 12 years or so, we all should be very grateful to have disruptors of their kind; otherwise commerce just moves at it's own convenient pace.
One of Lex’s better videos.
Had no idea Dax Shepard's older brother, Jim, was a microprocessor engineer.
This is one of the most funny videos I have ever seen in a nerdly like way
Doing something that is very difficult mastering it. Becomes easier and easier I like the way he say machines do not think like humans. He's basically saying if you work hard at it anything is possible.
Autonomous driving is kinda already solved in simulations. The main problem is gathering a precise model of the environment and predicting all the collisions accurately. It's a hard problem, but's not a "solving human intelligence" kind of problem. It's more of a balancing pole kind of problem. I think.