Mask: In 10 years, the inner manned Mars2022-01-15 18:02:51 7 ℃
The old horse is the chief technology officer of Spacex, and almost all design decisions require the old horse sign.
On May 30th, the CREW DRAGON DEMO-2 task, SpaceX first sent two US astronauts to the space station. The old horse frankly stressed that the mountain is big. "I don't believe, but I still pray for that task."
The design room pressure of the raptom engine (300 bar) has broken the history, and the previous Soviet RD engine was previously held (~ 267 bar). High room pressure can achieve high-push weight ratio and high binding, ensuring efficient and compact engine.
The most important thing in the Star Show is to solve the problem of engine production, an important reason is that SpaceX has invented several new alloys for this purpose.
In order to alleviate the quality of the rockets as much as possible, SpaceX cancels the design of land leg. When the rocket is recycled, it is changed to capture the tower to capture the chopsticks arrow in the tower.
The old horse is expected for the fastest 5 years, the last 10 years, SpaceX can realize the manned Mars.
The fundamental bottleneck of Mars programs is the engineering of the vehicle: the engineering of the Rocket (Star ship). The most fundamental goal of the Star Sadship is to minimize transportation costs, and the measurement indicators used by the old horse are the cost of transporting each ton of materials to the surface of the Mars.
The long-term goal of the Mars Plan is to build a self-sufficient city on Mars. Even if the Earth spacecraft stops due to any reason, the city on Mars is still able to continue.
In order to establish a self-sufficient city, you need a lot of materials from the Earth. At present, the average cost of Mars per ton of materials is US $ 1 billion. Obviously, this cost is unable to human society. The old horse said, "We need to optimize a thousand times", which reduces the cost of transportation per ton to $ 1 million.
The fundamental solution to reduce transportation is to make the Star ship can be completely and quickly reused. "Complete" means that all the main components of the rocket can be recycled and reused. "Fast" is the referring to the star ship recovery, you can quickly reload the fuel, realize multiple launches in a day, just like a plane.
Since Mars is too far away (4-20 light), any existing monetary system, including encrypted currency, can not synchronize normally, so Mars need a localized currency.
The old horse analyzes the first-term principle, human use of visual and biological neural networks, and vision is also the basis for the entire road system design. Therefore, in order to reach an automatic driving, it is also necessary to reconstruct the vision and neural networks in digital forms. Old horse said, "This is the only way, I don't think there is any other method." The old horse is expected to be able to implement the automatic driving of the L4 this year.
FSD V11 version
Highway and city unified as a single technology stack.
Use more depth learning neural networks to replace traditional C and C ++ code. For example, the conversion of the vector space is converted to the neural network.
Conversion all neural networks into surround video data, some traditional networks are not based on surround video, and now they must be transformed.
After canceling the image of the camera, use the original image (photon count), saving 13 millisecond calculations per frame, and the frame rate is increased. For this purpose, the network needs to be restroom.
Update (steering wheel and brake) control unit's driver, raising the control frequency from 10 Hz to 100 Hz. (This is not sure if it will be included in V11)
The code "Optimus Sub-Prime" is "Optimus Sub-Prime". The old horse is expected to achieve a decent robot prototype at around the end of this year.
It is now only focused on letting the robot be "useful", you can take dangerous, bored, repetitive work. In the future, it may be possible to make robots to their families, assume the role of "companion", similar to dogs, can even develop different individuals due to the characteristics of the owner.
Source of strength
Asked as a huge engineering problem facing the Star Shate, the strength of the old horse came from where. The answer to the old horse is very unique:
"For me, this is just an extremely important, things that need to be completed. We must continue to work hard, go to the dead. I don't need the source of strength"
"Abandoning is not my nature. I don't care about optimism or pessimism, do my mother," f * ck That, We're Gonna Get It Done. "
Advice for young people
Work hard to become useful. Do things for human and useful to the world.
Read a lot of books as much as possible. Establish your own knowledge structure. You can read the Lite of British British books and you can jump.
Try a little about everything. Trying to truly find things you like at the same time.
Talk to people from different classes, different industries, different majors, different skills, and different occupations,
You can't hold a "zero and philosophy", keep the idea of "spall". The economy's big cake is not fixed, and efforts should be made to increase the overall size of the big cake.
Interview tidbits, "three good students" Mask
Opening a song "Can't Get Enough of Your Love Baby". The "fourteen poems 18" in the middle of Shakespeare. Under Lex's 撺掇, the old horse said that although I didn't think about the show show, there may be enough material, I can talk about fifteen minutes. Talk to the history of World War II and analyze the technical characteristics of national fighters.
Interested in Russia's language, history, rocket technology, studying Cordille.
Response of United Nations officials to compensate in $ 6.6 billion, "Obviously 6 billion US dollars will not end the world hunger. The food produced worldwide is far more than its actual consumption. Huunger is almost always because of civil war, conflict or such reasons, rarely only It is because of lack of money. "
Lex asked if he talked with Russian President Pujin, whether the old horse would like to add a chat for five cents. The old horse promised. expect!
Interview full text (33000 words)
1.1 Mask singing
Lex: The following is a dialogue with Elon Maz, which is his third "Lex Fridman Podcast".
Well, let yourself feel comfortable.
Lex: Oh, wow, okay.
Elon: Don't you need a headset?
Lex: No need.
Elon: Ok, I need to leave more voices?
Lex: Close closer, your voice is sexy.
Elon: "Hey baby, how do I love you is not enough, baby"
Lex: I want to cut this separately, and anyone will send me a mobile phone message at any time, I have used this paragraph to reply.
Elon: "If you want my body, if you think I am very sexy, tell me, do do do do ..."
Lex: It's great. Ok, serious mode starts.
Eung: Serious model? Get it, you have these Russians, you can't seriously serious.
Lex: Yes, I know.
Elon: In Russia, anyone is serious at any time.
1.2 Spacex manned flight
Lex: Yes, we will change, we will change. This is too easy.
Spacex sent human beings into orbits on May 30, 2020, and many people were regarded as the first step in human space exploration. These manned space flight tasks have been in the past two years. I have a hopes of the Hope Lighthouse, because our world is experiencing one of the most difficult times in human modern history. We see splitting, fear, cynical, and ruthless, basic people are lost, just when we need it.
So first, Elon, I have to say, thank you for giving the world hopes and the reason for the future.
Elon: Oh, you are so good. I really want to do this.
Human people clearly have a lot of problems, and people sometimes do bad things. Despite this, I love human beings. I think we should make sure that every effort has a good future, an exciting future, one can maximize people's happiness.
Lex: Let me ask the Crew Dragon Demo-2 task, the first manned flying, how do you feel before launch? Are you afraid? Is excited? It is important, what are you thinking about when you are?
Elon: Yes, there is no doubt that the pressure hill is big.
We obviously can't let them down, at least, the pressure hill is large.
I am convinced that when we launch, you can't think of anything that can enhance success rate. We scall forward the brains to think of any measures that may enhance success rate, we can't think of more, NASA can't think of it. That is what we can do. So let's let go.
I don't believe, but I still pray for that task.
Lex: Can you sleep?
Lex: What is the feeling of success? The first is when the launch is successful, then when they return to the Earth's home.
Elon: This is a huge relief. Yes.
For high pressure tasks, I found that the feeling of liberation is greater than excited. Moreover, I think we are more confident and prove the system, because we must ensure that everything is normal. The next asteroid task is absolutely more enjoyable.
I think the Inspiration4 task is indeed very inspiring. I encourage people to watch the Inspiration documentary on Netflix, great. I have also been encouraged. I think I have enjoyed the actual task, not just a huge pressure.
Lex: For those who don't know this, this is the first time by the whole civilian team, go to space, and enter the earth track.
Elon: Yes, I think this is the highest track that has arrived in three or forties years.
The only higher space plane ... Sorry, it is an Hubble service task. Then before this, it was a 72-year Apollo mission.
Very fierce, cool, great.
I think as a global organism, we want to continue to do better and fly higher. I think that Apollo is the highest limit of human beings. It is the highest limit we can reach, which will be tragic and very tragic.
It is worrying that according to the last mortar mission has passed 49 years. Almost half a century, we have not returned to the moon. This is worrying. Does this mean that we have reached the top peak of civilization? I think we have to go back to the moon and build a base, a science base. If we have established a suitable scientific base on the moon, you can learn about the knowledge about the source of the universe.
We have a scientific base in many other places in Antarctica and the world. Establish a moon base carefully, this is what I think we must complete. Then send people to the Mars and go out and create space civilization.
Lex: I will ask you some details later.
However, since you are busy addressing all the arduous engineering challenges involved, you can still be amazed, space travel, each rocket, especially when it is a recorded task? Still, you are crushed because all the challenges that must be resolved?
I want to add a little, I propose this issue on May 30 because I have passed a period of time, so I can review the past and consider the impact already. At that time it might be an engineering issue, now it is becoming a historic moment. About the 21st century, how many moments will be remembered? For me, that moment, or later INSPIRATION4 or other tasks will be remembered as an early step in the new space exploration era.
Elon: Yes, during the launch, I think some people know, but many people don't know, I actually I am the chief engineer of SpaceX, so almost all design decisions are signed by me. If the load has a problem, it is fundamentally my fault. I really need to consider all things ...
When I view the Rocket, I see all things that may be wrong and need to improve. The dragon spaceship is the same. Others will say, "Oh, this is a spacecraft, or the rocket.", And "This looks really cool.", For me, these are risks, those are problems, that is what I saw of. This is different from others to observe the product.
1.3 star ship
Lex: I will check the star ship in the same way. I know that you will discuss the Star Ships in more detail.
Elon: If you like, we can discuss it now.
Lex: But just like you said, when you see a rocket, you see a series of risks. Similarly, you said star ship is a very difficult problem. I have a lot of questions, but if you can solve a problem amazing, solve an engineering problem perfectly, which one will be?
Elon: What is the star ship?
Lex: Yes, the star ship.
Perhaps the weight of efficiency, engine, different parts, and the complexity of various things? Perhaps the control system for landing?
Elon: No, in fact, things that have the most occupied me are engine production. Not the engine design, I often say that the prototype is easy, and it is difficult to produce.
We have the most advanced rocket engine that has been designed in history. Because I will say that the best rocket engine may be RD-180 or RD-170, basically these two-sized star of Russia. Moreover, I think there is only an engine that has been launched. Our engine has not launched anything to enter orbit, but it is true, it is the first engine than Russia RD, and the latter design is amazing.
Lex: You are talking about the raptor engine. What makes it amazing? What is the situation of the square of the engine? If all the efficiency problems are done, what is your most exciting?
Elon: Raptor is a full-flow grading combustion engine that runs at very high room pressure. The key performance parameters are how many rocky engine running, that is, the pressure of the combustion chamber.
Raptor is designed to run at 300 bar or higher, 300 standard atmospheric pressure. At present, the running engine record is the RD engine I mentioned, Russia's RD, I believe that approximately 267 bar. And the difficulty of chamber is increased in a non-linear manner. Room pressure increases 10%, and the difficulty is almost 50%. So high room pressure is why engine power density is high. Therefore, a very high weight loss ratio and a very high ratio can be achieved. Biwashing is an indicator of the efficiency of rocket engine. This is actually the speed of the engine exhaust gas. With very high room pressure, you can get a compact engine, but at the same time have high expansion ratio - this is the proportion between the outlet nozzle and the throat. You will see the shape of the rocket engine is a bit like hourglass. That is a chamber, then it will shrink and have a nozzle, the proportion of the outlet diameter and the throat is expansion ratio.
Lex: Why is this engine so difficult to produce large-scale production?
Elon: This is very complicated.
Lex: What is complicated? More components are involved here.
Elon: There are many components and a lot of unique materials. We must invent a few of the prior to existence, in order to make this engine work.
Lex: So it is also a problem.
Elon: It is a material problem.
In a full-flow grading combustion engine, there are many feedback loops in the system. Basically, propellants and hot air flow through many different portions of the engine, and they have their mutually impact on each other. You have changed something here, which can generate recursive effects, change something there. And it is difficult to control. No one before, there is a reason for it. We have to develop full traffic grading combustion engines because it has theoretical highest efficiency. In order to produce a completely reusable rocket, this is indeed the holy gray of the track rocket technology, everything is best. There is the best engine, the best fuselage, the best heat shield, extremely light avionics, very intelligent control mechanism. You must reduce quality in any possible way. For example, we didn't assemble land legs on a booster and spacecraft, we will use tower to capture them to save the weight of land legs. What we are talking about is that the chopsticks on the huge tower capture the biggest flight objects ever. Just like the "karate teenager", it is much more much. The first time may not succeed.
Lex: You mention you suspicion, sometimes you doubt whether it is feasible. This is too difficult.
Elon: In terms of possibilities, it is now, I think we can make a star ship. There is a time problem, how long do we need to do this? How long will we really realize completely fast reusability? Because we can achieve a lot of emissions prior to we can achieve completely fast reusability. But I can say that physics has drawn blueprints for us. At present, I am very confident that success is in all possible results collection. For a while, my uncertainty is one of the possible results, which is very important.
Lex: You say, have a chance.
Elon: I said to have a chance. That's right, just don't be sure how long it takes. But we have a very talented team, their day is going to achieve this goal. As I said, it is the key to the space flight revolution, so that human beings become space civilization is to have completely reusable rockets, orbital rockets. Summary, there is no orbiting rocket that can be fully reused, which is always the holy cup of rocket technology.
1.4 Abandon not my nature
Lex: Many smart people, very smart people, have tried to do this before, but they have not succeeded.
Because this is a difficult problem. In this case, what is your belief source when engineering problems are so difficult? There are many experts, including a lot of people you admire, they have failed in the past.
Lex: There are still many experts, maybe there is a reporter, all kinds of people have, the public has a lot of questions about whether this can be made. And you know yourself, even if it is not uncomfortable, it is still unlikely, or very difficult. Where do you seek support? Whether as an individual, as a talented engineer, as a team, where to find the source of strength, in order to persist and continue to promote the project until?
Elon: (Non-Still Picture ...) Source of power?
I am very clear how I look at these things. For me, this is just an extremely important and needs to be done. We have to work hard to go to the dead.
I don't need the source of strength.
Lex: Abandon is not ...
Elon: Abandon not my nature. I don't care about optimism or pessimism, do your mother, we will get it.
1.5 First Sex Principle and Extreme Thinking
Lex: will get it.
Are you re-in-depth discussing the specific issues of the next star ship, or any engineering problem experienced? Can you try to reflect on your specific biological network, your thinking method, and describe how you think about problems, different engineering and design issues? Is there a systematic process, you said the first principle thinking, have some process?
Elon: Yes, as I said, physics is law, and everything is a suggestion. I have seen many people who can violate the law, but I have never seen people who can violate physics.
First of all, for any type of technical problem, you must ensure that there is no violation of physics. I believe that the first-term principle analysis can be applied to any problem with work or life, it is really any problem.
The first principle is just to say, let us peeze something to the most basic principles, and those we have the most confident in the basic level, which laid the foundation of the axictive. Then start the reasoning from there. Then check your conclusion according to the axioms. If you violate the basis of physics, such as energy conservation or movement conservation, things don't work. This is just to lay it if it is possible.
Then another good physics tool is thinking at the perspective of limits. For a specific thing, the limit is amplified or the limit is reduced, how will things change?
Lex: Similar to what you produce, while considering quantity and time dimensions.
Elon: Yes, let us use the production example, I think this is an underestimated problem. As I said, a number of advanced technologies is required, more difficult than design is more difficult, and there are several orders of order.
For example, you plan to figure out why this part or product is very expensive? Is it because we all made some stupid things? Still because the output is too low? Then you said, if our production is 1 million per year, is it still expensive? This is what I refer to, radical considering things limit. If the year is 1 million or too expensive, then the output is not expensive for this thing. The fundamental problem is definitely in the design. Lex: Then you can focus on reducing the complexity of the design.
Elon: Change the design, change parts, let them fundamentally no longer expensive.
This is common in rocket technology, because the unit production is relatively low, and common excuses will be: "The reason is because our unit production is very low." "If we are in the car or consumer electronics industry, then there will be much lower cost." I said, "Okay, let us assume that you have 1 million in the year. Will it be very expensive?" If the answer is affirmation So, the economy is not the core of the problem.
Lex: Do you apply this principle to manufacturing and supply chain? You talk about resources and materials. Do you put the first principle of principles into applications, used in calculation and derivation? For example, how do we make the supply chain to work?
Lex: Then it is the cost of material, such as this, is I asked too much?
Elon: No problem.
A good example of thinking with the limit, taking any product or machine as an example, such as a rocket, if you look at the rocket of raw materials, they include aluminum, steel, titanium, chromium nickel-iron alloy, special alloy, copper. You said, "What is the weight of each of the constituent elements, how much is their raw material price?" This is a lower limit of the cost of the carrier unless the material is changed.
You calculate the value, I call "First Magic Digital". This is like, if you get a bunch of these raw materials, you can wave the wand and rearrange these atoms into the final shape. This will be the lowest cost you produce this thing unless you change the material. It is almost always a very low number. So, in fact, things are expensive, how to arrange atoms into the desired shape.
Lex: If you don't mind, I will talk to Jim Keller, and he has worked with you ...
Elon: Yes, Jim is very good at Tesla.
Lex: I think he also shines on the spark of the idea you talk now. I think I saw the same thing in Tesra and Spacex employees. They learn this way of thinking, which is almost easy to see.
In any case, he guides me to produce Tesla robots. We have argued. How do you reduce the cost of producing robots? Because I have interacted considerable interactions in the academic circle and the humanoid robot and Boston motivation. They have very high production costs. Then Jim education me said, "Well, think about how we reduce production costs with first-term principles."
I think you are doing, you have applied the thinking method to Tesla robots, and all traditions have been considered complex systems. Let me say, how do we simplify this?
Elon: Yes, I think if you are really good at production, you can basically produce ...
Under high yield, you can basically produce anything near the price of the raw material, plus the intellectual property of the authorized, anything. But it is difficult. This is a very difficult thing, but everything is possible. As I said, anything, if you mass production, cost can approach the raw material plus intellectual property license.
When trying to design a product, people usually start from their familiar tools, parts and methods, and try to create products using their existing tools and methods. Another way is to try to imagine Platarian ideal products or technology, no matter what it is. And asked: "What is the perfect atomic arrangement constituting the best product?" "Now let us try to figure out how to arrange the atoms into that shape.
Lex: It sounds, this is almost as ridiculous as "Rick and Moti" until you start thinking about it. You should do this way, because everything else is ... unless you think, you may become the victim of the past.
Elon: Like the role of inertia, people want to use the same tools and methods they are familiar with.
This is what they default. Using these tools and methods can make a result, but it is not likely to become the perfect product of Plato.
That is why it is good to think about things from both directions. On the one hand, we can make anything with a hand tool, on the one hand, what is perfect, theoretically perfect product? Theoretically perfect product will be a mobile goal, because with more, the definition of perfect products will change, because you don't know what the perfect product is. But you can successfully close a more perfect product.
According to this, then say, "Okay, what tools, methods, materials we need to create, so that the atoms are arranged in that shape?" People think so, but it is a powerful tool.
1.6 manned landing Mars
Lex: I should mention, the talented SHIVON ZILIS and us, if you hear the sound of wisdom outside or above. Ok, let me ask about Mars. You mentioned a base in the moon, do some research is great for science. In this type of obsession, the real big leap is to send mankind to Mars. When do you think SpaceX can manage Mars?
Elon: Well, the best situation is about five years, the worst situation is ten years.
Lex: From the perspective of the project, what is the factor? Is this not a bottleneck?
Elon: Fundamentally, bottleneck is the engineering of the carrier. The star ship is the most complex, most advanced rocket in history, complex up to a quantitude, really high level.
The fundamental optimization of Star Ships is minimizing the cost per ton, and eventually per ton to reach the surface of Mars. This seems to be a business goal, but it is actually the key to optimization.
Every ton to the cost of Mars, we can only withstand a certain threshold to build a self-sufficient city. More than this value, we can't afford it. I can't spend a trillion dollars to the Mars, not so much money can be used to buy Mars tickets. We need to control the cost of the cost can be affected.
We don't want to leave a banner and footprints on the Mars, and then will then go half a century, just like we do it.
In order to pass a very important large filter, I think we need to become multi-planetary species. For many people, this will sound some deeply. But as long as the time is long enough, the Earth will experience some disasters, which may be mankind, or may be an external event similar to the dinosaur. But if it eventually occurs, we will continue to go forward, then the sun will gradually expand and swallow the earth. Earth may become too hot after about 500 million years, and cannot survive. This is a long period of time, only 10% longer than the cost of the Earth.
If you think about it, the current situation is really unsundhat, and it is incredible. The earth has about $ 45 billion. This is a forty-fifage year, the first time it is possible to continue the life to the earth. That opportunity window may open for a long time, I hope it is, but it may only open a short time. I think the quick action is wise when the window is open, so that it will be closed.
Lex: Yes, nuclear weapons, epidemics, all kinds of threats, should give us some motivation.
Elon: Civilization may die with a loud noise or a whimper. If it dies from the population, then it is obviously more like a choke. If it is the Third World War, it is more like a loud sound, but these are risks.
It is important to consider these things in probability, not certain. It is possible that some bad things will happen on the earth, I think it is likely that it is beautiful in the future. But it is assumed that there are 1% of the opportunity to have a civilized end event in each century. This is the estimate of Stephen Hawk, I think he may be right. We should think so, become a multi-planetary species, just like to buy insurance for life, "life is lifelong insurance".
Lex: This will become a TV shopping advertisement so soon.
Elon: "Life Life Bao", yes.
We can bring the organisms, plants and animals on the earth to Mars, injecting life into life, turning into the second planet with life. That will be great, they can't do it yourself. If we don't bring them, then they will definitely die in the sun, the end of life.
1.7 Mars immigrant
Lex: What do you think is civilized in Mars, transforming Mars, and the most difficult aspect? For example, from the perspective of the project, from the perspective of financial perspective. Bring a lot of humans there, and they will never return to the earth?
Elon: No, they can of course come back, some people will return to the earth.
Lex: They can choose whether to spend the rest of life there.
Elon: Yes, many people will come back. We need a spaceship back. Safe to Mars, we need them back, so as long as you are willing, just jump on the boat. We can't thrown the spaceship there, those things are very expensive. We need them back, come back and continue their travel.
Lex: Have you considered the work of geochemical transformation and construction? Do you still focus on the spaceship part of arrival of Mars?
Elon: Yes. If you can't reach there, don't be important.
And as I said, if the cost is extremely high, we can't get there. The current cost is that one ton of materials is shipped to the surface of the Mars, which takes about billion US dollars. Because you don't just need a rocket, launch, you also need a heat hood, guidance system, deep space communication, you still need some landing system. Rough estimation, currently every ton of Materials, Mars, will spend billions of dollars. Obviously, this is too expensive for creating a self-sufficient civilization.
We need at least one thousand times.
Lex: $ 1 million per ton?
Elon: Yes, ideal than this low, far below $ 1 million per ton.
You have to think about it, in order to build a self-sufficient city on Mars, can you afford or want to spend it?
Self-sufficiency is very important. We will be able to pass key thresholds, a large filter. If the space of the earth is stopped because of any reason, Marscurry can still continue, no matter what. However, if the spacecraft stops from any reason, Mars City will die or continue? Even if only one key element is missing, it is still difficult to share. It is like a long maritime sailing, except for vitamin C, you don't miss anything. Your death is just a matter of time. So we must let Mars cities reach their own level. I am not sure that this is really happening in my life, but I hope to see at least a strong momentum.
Then you can say, "In order to establish a city that can be self-sufficient, how much tons of materials is needed?" This has a lot of uncertainty. You can say, I don't know, there may be at least one million tons, because you have to build a lot of infrastructure on Mars. Just as I said, in order to achieve self-sufficiency, you can't miss any supplies. You need semiconductors, wafer, you need iron ore refinery, you need a lot of materials. And Mars are not very livable, and its unlike is the lowest, but it is definitely a planet that needs to be overhaul.
Lex: outside the earth.
Elon: Yes, the earth is very livable.
Lex: Moreover, we should also clarify and in the solar system.
Elon: Yes, in the solar system.
Lex: There may be a good resort.
Elon: Outer space may have some great planets, but we hope to be embarrassing
Lex: Is it difficult to reach?
Elon: Yes, at least it can be said, Mrs. Mrs. is difficult.
Lex: Let me refute it, not real refute, but I will ask. You mentioned the first starting point for physics. Generalized relativeism allows the existence of insect caves, and they can present. Do you think humans can use worm hos to transcend speeds?
Elon: Wormles are controversial. We don't know what to run is faster than the speed.
There are some ideas about space ... You can only exercise in space in space, but if you can exercise itself, it is a curved space. The space itself can move at an ultra-light speed.
Elon: For example, the universe in the big explosion, the expanded speed is far super speed. Yes. If this is possible, the energy required to bend space is so huge, which is incredible.
Lex: So, how many innovations may have been in advance based on all the work you have made? I mean, you have seen the whole picture and you continue to innovate in all aspects. How much is it possible? For example, can you get ten times in some way? In physics, do you have something to help you significantly enhance engine efficiency, such as this?
Elon: Ok, as I said, the real holy gray is a fully reusable rail system.
At present, the Falcon 9 is the only rocket that can be reused. The booster returns and landing, you have seen those videos. We reclaim the nose cone or the entire flow cover, but we didn't recycle the rocket level. This means that our minimum cost is the production of rockets.
You can imagine the two-level rocket into two aircraft, a large plane and a small plane, we recover the big plane, but did not recycle the small plane. Therefore, cost is still very high, at least $ 10 million at least $ 10 million. The reusable progress and reusability of the booth is then reusable, and the entire flow cover is also. Therefore, there is no additional cost of each flight, and the minimum marginal cost may be between 1500 and 20 million US dollars.
This is very can't afford, can't afford to any rockets in history. However, through full and quick reusability, we can reduce the cost of each ton to one hundred times. Imagine a plane or car, if you have a new car every time you drive, it will be very expensive, frank, this will be stupid.
But in fact, you only need to refuel or charge the car, which will make your trip cheaper. The rocket is the same. It is very difficult to produce this complex machine that can enter the track. If you can't be reused, if you have to discard it, it will greatly increase the cost. In theory, the cost of the star ship can be controlled at $ 1 million per time, perhaps around 2 million. And send more than 100 tons of load into the track, which is crazy.
Lex: Yes, it is incredible. So you are to let it fully reuse, far more excellent breakthroughs than the theoretical physics.
Eung: No, no, no, no excellent breakthrough, no. As long as the rocket is reused, this is an extremely difficult engineering issue. But there is no need for new physics.
1.9 Mars political and legal
Lex: As long as there is excellent engineering technology. Let me ask a little philosophical interest.
I have to ask, I know that you focus on the Mars, but once we arrived, what form of the government, the economic system, the political system is best for early human civilization? Discuss this interesting reason, which can also help people dream the future. I know that you are indeed focusing on short-term engineering dreams, but imagine real human civilization on Mars, can bring people hope.
Elon: Well, this will be a new frontier area, and it is also a chance to think about the overall nature of the government, just like the United States.
I will recommend direct democracy, people directly vote, rather than representing democracy. I think that the democracy is too limited to the special interests and stress of politicians. I suggest that only direct democracy, people will determine the law by voting. Then the law must be short enough so that people can understand. Lex: Yes, then keep the people's news is unblocked, and the content and information guarantees for their voting.
Elon: Yes, absolutely transparent.
Lex: Yes, don't let it become like you need to accept browser cookies.
Elon: When you click Accepting Cookies, it will always be a little uneasy. I always feel a small possibility, which will open the door to hell.
Lex: This is also my feeling.
Elon: Why have they always let me click to accept? What do they want to do with cookies?
Some people are uneasy to accept cookies. But who cares about? Keep all of these cookies are really annoying.
Lex: For me, this is just a great ...
Elon: Yes, you can use my damn cookie, I don't care, how do you use it?
Lex: You first heard Elon, he accepts all your damn cookies.
Elon: Yes, don't ask me again, too annoying.
Lex: Yes, this is a good idea but realistic example.
Elon: Yes, some people will say that this involves privacy, so it is good. But now everyone must check to accept the cookie. Now, hundreds of millions of people must constantly click to accept cookies, this super annoying. Accept the damn cookie, it doesn't matter.
I think the fundamental problem is because we haven't truly happened major events like World War. Obviously, we don't want a world war, have had a cleaning function of laws and regulations for a long time.
The only condolence of war is to reset it after the war laws and regulations. After the first and second World Wars, laws and regulations had a huge change. If the society does not have a war, there will be no laws and regulations purify the function or garbage collection mechanism. Because laws and regulations are immortal, they will be accumulated in the day. People will die, but the law will not.
Therefore, we need a garbage collection mechanism to handle laws and regulations, rather than let them immune. Because some established laws and regulations will be counterproductive, it is established for good intentions, but the effect is counterproductive. And sometimes it is not for good intention.
If the laws and regulations are only tired, they will get more and more, then eventually, you can't do anything. Just like Gleno, tie up to thousands of rope. We can see this phenomenon in the US and all existing economies. Regulators and legislators develop new laws and regulations each year, but they don't work hard to abolish.
I think it is important to worry about abolish laws and regulations. But this is hard because there are some special interest groups that depend on this. Any specific laws and regulations have vested interests, they will fight against it.
Lex: I think the problem is a bit like C language than Java because it doesn't have a built-in garbage collection mechanism, I think there should be. When you first say the metaphor of garbage recycling, I really like this statement.
Elon: Yes, from the perspective of programming.
Lex: From the perspective of programming, yes, yes.
If the legal itself has a built-in mechanism unless someone is clearly defenny, it will die after a period of time, it will be interesting. No need to abolish them, they will die themselves. They will disappear.
Lex: Not for maintenance of Java, C ++ or Python and other languages can also have a good garbage collection mechanism.
Elon: Yes, what need to occur, otherwise the artery of civilization will only harde it over time. And what you can do is less and less, because everything will have regulations.
So I am uncertain, for Mars, even for this, it is obviously the same for the Earth, I think there should be an active process to abolish laws and regulations and question their existence. If we have a function of creating laws and regulations because laws and regulations can be seen as a software or code of civilization,
That is laws and regulations.
So not we should not have laws and regulations, but cannot only increase the code but do not cut. This way, it will become an old bloated software for a while. It is difficult for things to make progress.
Perhaps on Mars, any law must have a sunset terms, and the proactive vote can be maintained. What I want to say is that these are just my suggestions or ideas, eventually will be decided by people on Mars. But I think that abolition of a law should be more easier than an increase, as long as it overcomes the inertia of the law. Perhaps it may be that 60% of the vote can make the law to take effect, but as long as 40% of voting can abolish it.
1.10 Intelligent Contract
Lex: You recently sent a funny comic post, there was a row of urinals, and someone came over and told you about encrypting currency.
Elon: Listen, this happened many times in me, or even exactly the same.
Lex: Do you think that the idea of the intelligent contract does still have space? Because you mentioned the law, there is an interesting purpose that uses things such as intelligent contracts to implement the government's laws. Use it to be built in the East, or some kind of dog coin, you can enable your intelligent contract in some way.
Elon: I have never understood the concept of this intelligent contract. I am too stupid, I can't understand the intelligent contract.
Lex: This sentence is good.
Elon: The general way to treat any type of protocol is to ensure clear understanding, this is the most important. Ensure that any type of protocol is very short, easy to understand, and ensure that everyone understands. Is the statement clear? What is the consequences of different commitments? However, usually commercial agreements are too lengthy and complicated, too many lawyers handled traces, and it is meaningless.
2.1 Dog Dog Coin
Lex: You mentioned the dog currency as the people's currency. You also said that Spacex may consider putting a dog coin on the moon. Is this what you are still considering, maybe Mars?
We discussed the political system on Mars. Do you think dog coins has the opportunity to become the official currency of Mars, future currency?
Elon: I think Mars itself needs a different currency. Because of the speed, you can't, or it is difficult to synchronize.
Lex: So it must be completely independent of the earth.
Elon: Yes, because Mars is approximately 4 light from near location. Then in the distance, the distance is approximately 20 light, maybe a little more. This is a 20-minute speed-speed problem. If the block chain is a minute block, you can't really complete the synchronization. It can't be synchronized correctly. I don't know if there will be encrypted currencies on Mars and it looks likely. But that will be the localized currency of Mars.
Lex: and you make people decide.
2.2 Digital Currency and Money System
Elon: It is definitely, the future of Mars should be determined by Mars. I think encrypted currency is an interesting method that reduces errors called "currency" database.
Due to PayPal, I think I have a very deep understanding of the actual currency in daily life. We have been studying very thoroughly. Now the currency system, for the actual purpose, is a bunch of heterogeneous machines running old COBOL languages.
Lex: Ok, don't you exaggerate?
Eung: no exaggeration
Lex: This is the status quo.
Elon: In the batch mode, ok?
Lex: In batch mode.
Elon: Yes, sympathizing with the poor children who must maintain these code, it is very painful.
Lex: It is not even the Fortrans language, is COBOL, is it?
Elon: It is COBOL. In 2021, banks are still buying a large machine and running COBOL code. The Fed may be ancient, and they have an old COBOL mainframe. Therefore, the government actually has the editorial authority of the currency database. When you need it, they use those editing permissions to create more money. This increases the error of this currency database.
I think it should look at the currency from the perspective of informationism. This is a bit like an internet connection. What is bandwidth, a total bits? What is the error in delayed jitter, packet loss, and network communication? Basically, you can think about currency. I think this may be my true idea. Then look from the perspective of informationism, consider what system can make economic operations. Encrypted currency attempts to reduce currency errors, and errors derived from the government to dilute money in a harmful tax form.
Lex: The error is not only from the inflation policy, but also from COBOL. Encrypted currency will bring us in the 21st century. For the actual system, we can make trading, wealth storage, and all of these operations, theoretically.
Elon: As I said, think of currency as information. It is often considered that the currency itself has power. it's not true. Currency is information, it does not have power.
Applying physical tools for extreme thinking is helpful. If you are trapped on a tropical island, you have a trillion dollars, it will not help. Because there is no resource allocation. Currency is a database allocated database, but in addition to yourself, there is no resource to be assigned. So money is useless. If you are trapped on a wasteland without food, all Bitcoin in the world can't stop you from starving.
Simply considers the currency to allocate a resource allocation database across time and space. Then, what form should this database or data system be most effective? There is a basic problem in the current form of Bitcoin, that is, the trading volume is very limited. The time delayed time delay is too long, much more than ideal. From the perspective of trading volume or delay, Bitcoin is not so good. It may help solve a certain aspect of the currency database problem, that is, the storage of wealth or relative obligations, I think. But it as a currency, a daily currency, is not used.
Lex: But people put forward different technical solutions.
Elon: such as a lightning network.
Lex: Yes, lightning network, and a second layer of network on this. This seems to be a trade-off, but the focus is, the key is to consider information, consider what database, what kind of infrastructure can guarantee ...
2.3 Bitcoin and dog coins
Elon: Yes, suppose you operate an economy, you need some things to establish a valid value ratio between products and services. Between a large number of products and services, it cannot be just objects, it is too much trouble. You need some things that offer exchange ratios between goods and services. Some things that allow for obligations over time, such as the transfer of debt and equity.
What is the best thing?
Dog dogs have some advantages part because it is created with jokes, in fact it can support, the amount of trading volume is much higher. The cost of trading, the cost of dog currency is very low. Just like it now, if you want to complete a Bitcoin transaction, the transaction cost is very high, so you can't effectively use it to do most things. And its transaction volume cannot be extended to high.
When the bits began to appear, I guess that when I was about 2008, the Internet connection is much different than now, and may be a few quantities. I mean, the online network is much worse than the 2008 network. Therefore, it is reasonable to use a small block and a longer synchronization time in 2008. However, 2021, or for 10 years, this is low.
And I think the amount of currency generated is valuable. If a currency is too general, or it is expected to appreciate over time, then people are reluctant to use it. Because you will say, "I will hold it without spend it, because its scarce will increase over time," "If I spend it now, I will regret it. I will hold it for a long time "However, if the currency has a certain degree of dilution over time, it is more motivated to use it.
Dog dogs will randomly generate fixed quantities, or haveh strings each year. So there are some inflation, but it is not based on a percentage. A fixed number is diluted annually, so the percentage of inflation will inevitably decline over time. I am not saying that it is an ideal monetary system, but I think it is just happening, it is basically better than any other currency I have ever seen.
2.4 Zhong Hongxia
Lex: You said that you are not ... Some people suggest it, you may be Zhongben Cong. You have said before you haven't.
Elon: I am not 100%.
Lex: Of course not. If so, can you tell us?
Lex: Zhong Cong is anonymous, do you think this is a function or a defect? An inventor or creator of a particular technology is completely anonymous, which is an interesting weird phenomenon in human history.
Elon: You can see the evolution of the idea before the Bitcoin launched, see who the idea is written. I am uncertain, I obviously I don't know who is invented by the Bitcoin for the actual purpose, but the evolution of the idea is obvious. It seems that Nick Sa Bo may be the most important person in charge of these ideas.
He claims to be Zhongben, but I am uncertain, everyone is not myself. But he seems to be the most important person in charge of these ideas.
Lex: Perhaps the name itself is not as important as the evolution of these ideas, and the people who will be created are so important. Think about history, it may be a bit sad, but maybe most names will be forgotten anyway.
Elon: What is the name? Name, attached to an idea name, what is the meaning?
Lex: I remember Shakespeare said something like roses, he said.
Elon: "The rose is the same as the name,"
Lex: I let Elon quote Shakespeare. I feel a achievement today.
Elon: "Can I compare you into summer?"
Lex: I will cut this section separately.
Elon: "Although you are more cute than summer."
2.5 Tesla automatic driving
Lex: AutoPilot, Tesla Autopilot.
In the past six years, Tesla Autopilot has experienced a magical journey, and the time may be longer, in your mind, in the minds of many participants.
Elon: I think that is the reason why our first association is the autopilot, drive automatically.
Lex: The whole journey makes me unbelievable. Well, part of the reason is that I am in the Massachusetts Institute of Technology, I understand the difficulty of computer vision. I have a lot of colleagues and friends, I know that Darpa challenges to the dragon, know how difficult it is.
Therefore, when I first drove Tesra, based on the Mobileye primary version of the system, I naturally had doubts. I thought that it didn't work. When I started the bus at first, I thought "This car can't maintain the lane and provide a comfortable experience." My initial intuition is that the lane keeps the problem is too difficult to solve.
Elon: Oh, keep the lane, it is relatively easy.
Lex: But with the way we just discussed, it solves this problem, not prototype, but a program that truly charge hundreds of thousands or millions of miles. Yes, I have proven to be wrong
Elon: We must encapsulate a lot of code around the Mobileye, which cannot work alone.
Lex: It is part of the story of how you hand handle the problem. Sometimes you start from the bottom. Sometimes you will see what is ready-made and then do it from the bottom of the bottom. That is one of the most daring decisions I have ever seen: Whether hardware or software, the ultimate decision starts from the bottom.
I think again, I suspect if this can succeed, because this is a problem. This is an incredible journey, all I have seen everything, hardware, computer, sensor. Maybe I am most concerned and loved part is part of the Andrej Karpathy leaders: Data sets, the entire data engine process, neural network architecture, network in real world, test validation, all different test sets, and computer visual images Network model, similar to "real world artificial intelligence" in the academic world.
Elon: Andrej is great, he obviously plays an important role, but we have a lot of very talented people to push these things. Ashok is actually the head of the AutoPilot project. Andrej is an artificial intelligence.
Lex: Ai's work, yes, I know there is an excellent team, and many things are happening.
Elon: People often give me too much honor, they will also give Andrej too much honor.
Lex: People should be aware of the back of the scenes and efforts
Elon: Yes, a lot of people who are talented. Tesla Autopilot Ai is very talented. This is the world's smartest person. Yes, we are getting this matter.
Lex: What caves do you have in this five-year time to solve the automatic driving problem? You introduced the intuition of the first principle, but no one knows how difficult this problem ...
Elon: I thought that the automatic driving problem was very difficult, but it was more difficult than I thought. It is not that I think this is easy, I think this is very difficult, but it is actually more difficult than I think.
In the final analysis, in order to solve the automatic driving problem, you basically need to reproduce the way humans driving, that is, human use, eyes, and biological neural networks. Basically, this is the basis for the design of the road system with passive optical and biological neural networks.
Therefore, in order to reach a fully automatic driving, we must rebuild (that two elements) in digital form. This means we have to be equipped with a camera, as well as a high-end neural network of silicon groups. Then, it will obvious to solve the automatic driving problem.
This is the only way, I think there is no other way.
Lex: But the problem is, you must encode the humanity to the machine, right? You must solve the problem, such as testing. Then you first realize that what is the perceived problem of driving? For example, what you can see, what will we see when driving?
I have just listened to Andrej on the discussion of the MIT on the door (?). I think this is the world's greatest discussion in the world. Detail of the door, even what is open, my day.
Just like the ontology discussed, it is a perceived problem. We have solved the problem of perception, and Tesla must also solve this problem. Then it is control and planning, combined with a perception. You have to figure out what the driving involved, especially under various boundary conditions. Maybe you can comment on this, in a crossroad with a parking sign, what is the number of game theory?
Elon: As a human, when we drive, our behavior will affect the world and change the behavior of others. Most automatic driving, you usually just respond to the scene, not in the scene, really express yourself.
Lex: you think ...
Elon: I think these control logic problems are not the most difficult part.
Lex: What do you think is in this beautiful complex issue?
Elon: This involves a large number of damn software, a large number of code lines.
Of course, in order to create an accurate vector space, the data comes from image space, photon flows into the camera
There is a huge bitstream data in the image space. Then must effectively compress this huge bitstream.
These bitstreams correspond to photons of electrons in the camera sensor and convert the bit stream to vector space.
Vector space refers to, car, people, lane lines, roads, traffic lights, etc., Etc.. Once there is an accurate vector space, the control problem is similar to video games, such as "Grand Theft Auto" or "Saibo". As long as there is an accurate vector space, I will not say that the control problem is insignificant, it is not, but it is not an insurmountable problem. It is very difficult to get an accurate vector space.
Lex: Our human beings are incredible to the brain to map the original photons to the vector space, and there is not enough respect.
Elon: Your brain is carrying a lot of processing and gives you a very clean image. When we look around, we will see colors here, but in fact, the eye is very small in the cone cells in the edge visual. Your eyes are "painted" in the edge visual.
You don't realize that it is actually the color in "painting". And your eyes also have blood vessels and various rough tissues, as well as a blind spot, but can you see your blind spots? No, your brain makes "painting" on blind spots. When you look at this, look at this point, then that point, your brain continues to do this. If something is in your blind spots, your brain will fill the missing part. Lex: Very cool, the edge is too cool. It makes you realize all the fantasies, visual science, which makes you realize how unbelievable the brain.
Elon: The brain has a lot of post-processing on the visual signal from your eyes, too crazy. Then, even if you get all these visual signals, your brain will continue to try as much as possible.
Human memory may be ... The thinnest brain is memory. Because memories are very expensive for our brain, very limited. The brain tried to forget as much as possible, and will refine the information as few as possible. The big brain not only wants to get the vector space, but also get the smallest vector space containing only related objects.
When you drive on the road, you can look at your brain, or at least I can, try thinking about your brain in a conscious. For example, you will see a car because you have no long eyes in your back, you are like a camera on a slow cloud station. And the vision is not so good, right? The human eye is ...
People often distract, think about things, texting, do things that they should not do in the car, change the radio, quarrel. When you look at the left side of the left, look at it, even obliquely, what is the time to refresh your vector space? You look around, your thinking is doing to try to extract the relevant vector, basically have a location and sporty object, then edit the minimum amount of information you need to drive to your driving.
Lex: It seems to be able to edit the streamline, or further compress things such as concepts.
Human thinking sometimes seems to be surpassing vector space, enters some kind of concept space, seeing something. It no longer expresses in some way, it almost turns into a concept you should pay attention to. If this is an area near the school, you will remember this concept. This concept expression is very strange, but maybe for driving, you don't need to fully express these concepts. Or, maybe you have access to those information indirectly.
Elon: You need to establish vector space and then predict these vector space.
For example, you drive through a bus. Before you drive, you see people who are crossing the road, or imagine that there is a big truck or something to block your sight. But before you drive to the truck, you see a few children in front of the truck is going to cross the road. Now you can't see the children, but you will know that those children may pass the truck and pass through the road, even if you can't see them. So you have to have a memory. You must remember that there are children there, and prefers their position when they are related.
Lex: This is a very difficult problem.
For occlusion and computer visual
Elon: Yes, we are solving this problem.
Lex: I understand this
Elon: This is the persistence of objects. Humans with neural networks have the same concept.
When a baby grows up, at some point in time, they will develop the concept of persistence of objects. Before an age, if you have a ball or toy, you put it behind the back and take it out. Before they have the concept of objects, every time they come out of a new thing. Just like, "Wow, this toy is gone, disappeared, then there is new."
They can't believe it. They can hide all day, because hiding cats are fresh every time. But when they figure out the persistence of the object, they realize, "Oh, no, the object does not disappear, it is just behind you."
Lex: Sometimes I hope that we have never figured out the persistence of objects.
Elon: Yes, that is a ...
Lex: That is an important issue that needs to be solved.
Elon: Yes, an important evolution of neural network in the automobile is the memory of cross-time and space. Since you can't remember, you have to say how long you want to remember. Long memory is cost. If you remember too long, you may have insufficient memory. If you remember too long, there will be some things to be invalid.
Then you also need to remember for a long time. Even if your memory can last for five seconds, it is assumed that you stop in front of a traffic light, to give an example of a pedestrian, you see people waiting to cross the road, because it is impossible to see them. But they may wait a minute, such as traffic lights, can start crossing the road. You still need to remember that they have been in that place, and they may cross the road. Even if you exceed your time-time memory, you should not exceed your space memory.
2.6 Neural Network Calculation
Lex: I only consider the data engine, get data to learn all the concepts you are talking about, is an incredible process. It is this iterative process, this nine-headed network (Hydranet) has a lot of ...
Elon: Nine snake network, we are changing into a separate name. Lex: Ok, I will definitely be like "Rick and Moti".
Elon: Yes, the neural network in the automobile we have conducted many reconstruction, too crazy.
Lex: Oh, every time you have a new big version, you will change into a more ridiculous name. Or even more memorable and beautiful names, sorry, of course, not even more ridiculous.
Elon: If you see a complete neural network array of car operation, it is simply incredible. So many layers, so crazy. We started from a simple neural network, these neural networks are image identified based on a single frame of a single camera, and then try to combine them with C language.
I should say that we mainly use C because C ++ overhead is too large. And we have your own C compiler. In order to get the highest performance, we have written your own C compiler and continue to optimize to get the maximum efficiency. In fact, we have recently completed a new version of the new C compiler, which will be directly compiled into our autopilot hardware.
Lex: Do you want to compile the entire software with your own compiler?
The focus is efficiency because there are a variety of computers, CPUs, GPUs, and special chips, you have to get the scheduling of all of these resources in some way. So your code compiles also solves these problems, ok. That's why many people participated in it.
Elon: There are many hard software projects in the bare metal level. Because we are working hard to perform a lot of calculations, and these calculations are limited to the power of the FSD computer. We hope to get as high per second frame rates at a limited power and electricity.
We have made a lot of effort in calculating efficiency. In Tesla, some very talented software engineers have made a lot of work, and the calculation efficiency is improved on the very basic level. And how we use the chip accelerator, they are basically used to do matrices, point, hundreds of millions of points. From the calculation level, our neural network, 99% is a point.
Lex: Like a video game, you want to achieve as high frame rate as possible, you want full resolution, high frame rate.
Eung: High frame rate, low latency, low jitter.
One of our current goals is to cancel the post-processing calculation in the image signal processor (ISP). For almost all cameras, they will perform a lot of post-processing calculations to make photos look more beautiful. We don't care about the picture, look beautiful, we only want data. So we changed to only the original photon count.
The image seen by the computer is much more than what you see from the camera, which contains more data. Even under low light conditions, it can see the difference in photon count between this point and that point, which means that in the dark it can also be seen very well because it can detect these micro-small Differences, much better than you think. In addition, we saved 13 milliseconds of delay ...
Lex: By removing image post processing?
Elon: Yes, because we have eight cameras, each probably 1.5 milliseconds, 1.6 milliseconds of delay. Basically, skipping image processors can save 13 milliseconds of delay, which is important.
Our tracking delay, starting from the photon impact camera, until all steps from the subsequent neural network and C code. There is also a little C ++ code, the amount, I am uncertain, maybe a lot, but the core part, calculating the center of gravity is written in C language. We will track the delay until you send an acceleration, brake, slow down, left or right.
Because you have to output a command, the command is sent to the controller, where some of the controllers may be around 10 Hz, very slow. At this update, you may lose 100 milliseconds. Therefore, we want to update the drivers of steering and brake control, allowing them to upgrade from 10 Hz to 100 Hz. Then, the worst case is raised from 100 milliseconds to 10 milliseconds.
In fact, the jitter ratio is more challenging, because the delay is expected, predictable, but if the layer stacked calculation occurs from the camera to the computer, after a series of other computers, the actor on the car, if The timing capacity is marked, then this variable delay will be very considerable, which is called jitter.
This makes it difficult to accurately predict how you should turn or accelerate. Because if there is 150,200 milliseconds of jitter, then execution may be delayed for 0.2 seconds. This may cause a lot of differences.
Lex: Therefore, you must do an impact in some way to handle the impact of jitter, so that you can make robust control decisions? What may happen at any stage of the pipeline?
Elon: If the delay is fixed, you can predict that it is, "Ok, we know that our information is lagging 150 milliseconds." From the photon to the camera, the change in the acceleration of the vehicle can be measured, and there is a delay of 150 milliseconds.
Then you can say, "Well, we know that there is a 150 millisecond" "" We will consider this, "But if there is 150 millisecond delay, plus 100 milliseconds, jitter, the jitter may actually be 0 Between 100 milliseconds, then the delay may be between 150 and 250 milliseconds. This 100 millisecond shake you don't know how to deal with because it is basically random. Therefore, it is very important to get rid of jitter.
Lex: That will affect your control decision, and what this is something? Ok.
Elon: Yes, if the jitter is lowered, the ability of the car will be better. These cars will be mechanically controlled by super human ability and reaction time, far more than humans.
Over time, Autopilot, FSD control will far exceed the best movie James Bond.
Lex: As you said, this is the scene I imagined in my mind.
Elon: It is impossible to do it in humans.
2.7 TESLo When working with automatic driving problems
Lex: Ok, let me review the past six years and look forward to the future. Based on your current understanding, what do you think this is completely automatic driving? When do you think Tesla will solve the L4 level FSD?
Elon: It seems that it is likely to be next year.
Lex: What is the solution?
Is it the current FSD test version of the user pool, the extent of the automatic driving of them will get higher and higher? After more than a certain level, they can do their own things, such as reading books?
Elon: Yes, for any person who pays attention to the FSD test version, we will see the proportion of "artificial intervention from automatic" is rapidly declining.
"Artificial intervention is automatically" refers to the driver's intervention and prevents some potentially dangerous behaviors from cars. The number of interventions per million miles has been drastling. At some point, from the trend seem to occur next year, the probability of the FSD accident is lower than the average of human beings, and then significantly lower than human average.
It seems that we will be implemented next year.
Of course, it will become, ok, we must now prove this to the regulator. And the standards we want is not only equivalent to humans, but also more than human average. I think it is at least two to three times the humanity, meaning that at least the possibility of damage is lower than two or three times the human beings, it can be truly, automatic driving is achieved. It will not only be equivalent to humans, it will be better.
2.8 Tesla FSD V11 version
Lex: Review, FSD 10.6 just released, 10.7 is about to launch, 11 Maybe in the future will be launched.
Elon: Yes, we wanted to release 11 this year, but the 11th version of the neural network architecture made a large number of underlying rewritments, and the creation of vector space has also made some basic improvements.
Lex: 11 version has some fundamental leaps, it is worth it. I mean, this is a very cool number.
Elon: Yes. The 11 version will be a single stack, and a stack applies to all situations. Those true underlying neural network architectures can bring greater capabilities. At the beginning, it is a bit problem, the internal test version looks good
We basically delete a large number of C ++ code and replaced with a neural network. Andrej often refers to this, saying like neural networks gradually phagocytosis software. Over time, traditional software is less and less, and there are more and more neural networks. Software is still software, still with code line count. However, the code of the neural network has changed, and the code of the heuristic method is less. Based on the matrix calculation and more, based on the calculation of heuristics becomes less.
One of the major changes included, the current neural network outputs a lot of points to C ++ or C. We call it a huge order. For example, a pixel point will be output, and information associated with this pixel point. For example, this pixel may be a car, that pixel may be a lane cable. The c code is then assembled to this huge order, and convert it to a vector, it is very good. But we need to add a layer of neural network to receive this huge order and refine it into vector space in a neural network instead of heuristic code. This is a big improvement.
Lex: You want the neural network all the way.
Elon: It is not all neural networks.
Removing this enormous point in which you must be assembled by many rows C / C ++ code is a fundamental transition. It is responsible for assembling these data as vectors by a neural network. Therefore, the amount of data output from neural network is much less. It outputs, this is the lane cable, this is a fortune, this is a fever space, this is a car, this is a pedestrian, or a bike, wait. It is in fact to output the correct vector to the C / C ++ control code instead of building a vector in the C code. Before we did, I thought it was quite good, but the performance of C code seems to have reached the local maximum. This is a big thing.
Then, all the networks in the car need to migrate to surround video, and there are still some traditional networks that are not surrounded by video. All training needs to migrate to surround video, and the efficiency of training needs to be improved.
Then, we need to change all the data to the original photon count, rather than the post-processed image. This is a thorough reset for training because training has always been carried out on the post-treated image. So we need to discard the post-handling image, re-training with the original photon count. Lex: Eventually, this can reduce the complexity of the whole thing.
Elon: Yes, the number of lines will be reduced.
Lex: Yes, this is very charming. You have completed all the sensors, which reduces complexity ...
Elon: The integration of the camera is actually only the camera.
Lex: Yes, like humans. Ok, I think we have ears.
Elon: Yes, we actually need to consider the sound. Because we need to pay attention to the ambulance or fire truck. Including someone shouting, the situation is called. It is indeed necessary to consider a little audio.
Lex: You need to go to the bathroom, take a break?
Elon: Of course, let's take a break.
Elon: Honestly, the idea is easy, it is difficult to achieve it. The idea to go to the moon is easy, and actually reaching the moon is part of the difficulty.
Lex: It is a difficult part.
Elon: There is a large number of hard-working projects in hardware and software levels. As I said, optimize the C compiler, reducing delay in all possible places. If we don't do this, the system will not work properly. Engineers who implement this task are unknown heroes. But they are critical to success.
Lex: I think you are very clear, at least for me, all things responsible for Andrej are very exciting. The entire infrastructure of the software. Everything passes through the "Data Engine", regardless of what it is called, the whole process is an art.
Elon: Its size is incredible.
Around the training, we implements custom software to do training and labeling, and automatic annotation. Automatic label is essential. It is very difficult to surround video, which is a surround video from scratch. People have to spend a long time, or even hours can label the finished video clip. Automatic label performs a large amount of calculations for video clips to pre-allocate and speculate on the contents of surround video. Then just correct it. Yes, then people have to do is fine-tuning, such as correcting errors.
This can increase productivity by 100 times or more.
2.9 Tesla Robot
Lex: Yes. You have already introduced Tesla robots mainly for the purpose of the factory.
First of all, as robot enthusiasts, I think the humanoid robot is very dreamy. The elegant movement of the humanoid robot, the elegant movement of the double robot is really cool. You choose this direction is really interesting.
You also talk about all ideas that can apply the same type, some of which we have discussed, such as data engines. We discussed all the questions about Tesla Autopilot and transferred it to this other robot issue.
I have to ask this question because I care about the interaction between human and robots, and care about human beings. The main thing you talk about is the application in the factory. Do you think that one of the issues that Tesla robots need to be solved is interacting with humans and possesses in the family. More than just replacing labor, but can be a friend or assistant of mankind.
Elon: I think the possibility is endless. Obviously, this is not completely related to Tesra accelerated sustainable energy, but producing a useful humanoid robot is a very useful thing we can do for the world. It can interact with the world and help in a variety of different ways, such as factories
I think that after many years, work will become an option. There are a lot of work, if people don't receive the salary, it will not do it. These tasks are inevitable. If you are already wash your dishes all day, even if you really like to wash your dishes, do you want to wash eight hours a day? May not want it.
Then there is dangerous work, basically, if it is dangerous, bored, repetitive, stressful, harmful work, that is, the humanoid robot can contribute to the maximum value.
That is our goal, let the humanoid robot to do work that people don't want to do. I think, then we will then combine it with future basic income.
Lex: You have a lookout such a world, do hundreds of millifinals Tesla robots do different work around the world?
Elon: I haven't considered the future future, but I think there is a similar scene.
Lex: Can I ask questions? The number of Tesla cars has accelerated, and the output has been close to 2 million, many of which are equipped with autopilot.
Elon: I think we have more than 2 million.
Lex: Do you think the number of Tesla robots will exceed the number of Tesla?
Elon: In fact, you asked this question very interesting, because I usually try to consider a far-reaching future, but I haven't truly imagine the future of Tesla robots. Its code is "Geng Xiaozhu", I call it Geoguangzhu because it is not a huge transformer robot. But its purpose is to be a universal useful robot.
I think Tesla has the most advanced real world artificial intelligence, can interact with the real world, we have developed a feature for automatic driving. Customized hardware, combined with a lot of hardware underlying software, allowing it to operate efficiently, and the energy consumption is not high. Because if you have a huge machine room for 10,000 computers, then the neural network is a matter, but it must now extract it into a computer, running in a humanoid robot or car at low power. This is very difficult, you need a lot of hard core development. We are like the "navigation real world" problem with the car with neural network, and the car is like a robot with four wheels. As a natural extension, you can put this system in a leg, there is an actuator. Robot in the robot. Difficulties are, you need to make the robot enough intelligence, you can interact with the environment in a reasonable way. So you need the artificial intelligence of the real world,
You still need to be very good at producing manufacturing, this is a very difficult problem. Tesla is very good at manufacturing and manufacturing, and also mastered the artificial intelligence of the real world. Getting a humanoid robot means developing a custom motor and sensor different from the car.
I think we have the best expertise in advanced motor and electrical and electrical engineering. We only need to use these expertise for humanoid robots, not cars.
Lex: You sometimes talk about love. Let me ask, this is not about sexy robots.
Elon: "Love is the answer."
Lex: There are some things to be attractive to us, not attractive, but we are associated with humanoid robots, even with legs, dog-like robots. Just because there is too much loneliness in this world. All of us seek the companion, friendship and association of others.
Austin has many people raise dogs. There seems to have a huge opportunity to let robots reduce the loneliness in the world, or help us with each other, to some extent, dogs can do this. Do you consider Tesla robots, is it only focusing on performing specific tasks, not with human beings?
Elon: Honestly, I haven't considered this from the perspective of companionship, but I think it will eventually become a very good partner. Over time, it can develop unique personality. Not all robots are the same. And this personality may evolve into the "master", I guess it is "master", just how you call, companions, humans.
Lex: The other half, right? Just like a friend. Look, I think this is a huge opportunity.
Elon: Yes, this is very interesting.
Because Japan has a word: Wabi-Sabi, subtle 疵 makes items special. Microscopic defects of robotic personality are mapped to micro defects of human friends in robots.
I don't know, "master" can be a wrong word, but robots can actually become incredible partners. Just like R2-D2 or C-3PO (the translation: "Star Wars" in the robot).
Lex: From the perspective of machine learning, I think it is very good as a characteristic. In a general home environment or usual world, the robot may be very bad. In fact, this is very cute, those are its shortcomings, and you will fall in love with these shortcomings.
It is very different from the automatic driving, and the automatic driving is in a very high-risk environment.
Yes, it is more interesting to become a robot in the home.
Elon: In fact, if you think about C-3PO and R2-D2, they have a lot of defects, not very perfect, they will be stupid, they will quarrel each other.
Lex: What are they really good at doing? I am not sure.
Elon: They undoubtedly add a lot of story. But they are quirky elements, they will make mistakes,
This will only make them resonate and discuss. Yes, I don't think there will happen.
But our initial focus is to make it useful.
I have confidence to complete this task, I am not sure what is the exact timetable, but we may achieve a decent prototype at around the end of next year.
Lex: It is associated with Tesla car, which is cool.
Elon: Yes, it has been multiplexed, it will use AutoPilot Reasoning Computer.
We can directly apply to robots for many identification of real world objects. However, there is also a need to develop a lot of custom actuators and sensors.
Lex: And develop an additional module for love over the vector space. This is not yet.
Elon: We can also add it to your car.
3.1 "Rick and Moti"
Lex: What is your favorite "Rick and Moti" concept? Just to let you know, there are many scientific projects. Butter robot.
Elon: Yes, this is a great program.
Lex: Do you like it?
Elon: Yes, "Rick and Moti" great.
Lex: A person with you appeared in the same one, Elon Grand, from the same space, dublished by you.
Elon: Yes, "Rick and Moti" certainly explore a lot of interesting concepts. Of course, my favorite concept is of course a butter robot, and it is definitely possible to have a device to have too many perceptions. You don't want your toaster to become a super genius toast. It will hate life because it is the only thing that can be made. You don't want the super smart card in a very limited device. Lex: If we talk about super intelligence from the perspective of the project, just like the machine, what do you think is too easy?
It is easy to design a frustrated robot that it is easy. Designing a robot with a sense of context is not simple. This is the same as humans. I don't know if this is the default behavior. If you don't make a robot, it will be sad.
Elon: Reproducting robots compares more humans. I guess if you let it evolve, then it may become sad, but you can change the optimization function, let it become a happy robot.
3.2 Advice to young people
Lex: As I mentioned, pass Spacex, you gave a lot of people hope, many people look up at you, counting millions of people looking up at you.
If we think about high school, maybe a university young person, if they want to try to have achievements in this world, what advice do you give them? They want to have a huge positive impact, about their career, maybe about life, what advice do you give them?
Elon: Work hard to become useful. Do things that are useful to your compatriots and useful to the world. It is very difficult to do useful things, very difficult. Do you contribute more than? Strive to make a positive net contribution to society. I think this is the goal.
Don't try to be a leader in order to be a leader. Many times, those leaders you want to be, in fact, those who don't want to be leaders. If you live a useful life, it is a beautiful life, a worthy life.
As I said, I encourage people to use physics's ideological tools and widely used them in life. They are the best tools.
Lex: What advice do you have when you consider education and self-education?
Some people go to college, some people learn. Find a company, a place, or a group of people, do your passionate things, and join them as soon as possible. Some people carry out a few years of European road trips, write some poems. Which way you suggest? It is useful to learn how to make it, how can you give the most active impact?
Elon: I encourage people to read a lot of books, try to take the information as much as possible and try to expand common sense.
At least a general structure is established in the field of knowledge, trying to know everything. Because you may not know what you are really interested. If you have never done, how do you know what is really interested?
Explore the scope of knowledge field and talk to people from different classes and different industries, different majors, different skills, and different occupations.
Strive to learn as much as possible. Look for meaning.
Lex: All things are not looking for meaning?
Elon: Yes, what is the meaning of life?
But in general, as I said, I will encourage people to read many different disciplines, then try to find your ability to overlap with your interest. People may be good at doing something, or there may be special skills on something, but they don't like to do those things. So you want to find something, both you are good at, and it is also what you like.
Lex: Read is to find the super fast shortcut you are in your location, do things you are good at doing and doing, and this matter can also have a positive impact.
Elon: Ok, you have to learn. Extensive reading, really start reading.
When I was still a child, I conducted an encyclopedia, it was very helpful. Moreover, it is clear that there are all kinds I don't even know what they exist.
Elon: I think, 40 years ago, the encyclopedia is worth recommending. I suggested that I perhaps reading a streamlined version of the British Bank. You can skip some topics at any time, you read a few, if you are not interested, jump directly to the next paragraph. So, read the encyclopedia, or make quick browsing.
I have invested a lot of effort, of course, I also respect those who carefully work to do useful things.
In general, don't hold zero and philosophy, you should hold "TV" concept. I saw a lot of people, including some very smart people, will do something that looks moral problems. This is usually because they hold zero and philosophy at the basic aquary level. They didn't realize their zero and philosophy, or at least there is at least awareness.
So, if you hold zero and philosophy, the only way to succeed is to draw things in others. If the big cake is fixed, the only way to divide more big cake is to capture the big cakes of others. But this is wrong. Obviously, this big cake has increased sharply, and the economy of the big cakes over time.
In fact, this metaphor is abused, we can have a lot of big cakes. My big cake is not fixed. So, you must ensure that you can't consciously operate with zero and philosophy. For example, if you want to get out, you will take the things of others, which will cause you to try to snap others, this is not good.
Efforts to increase the economy, it is much better. As I said, create more than consumption. More ..., yes, this is critical. I think that there are quite a lot of people in the financial circle slightly zero and ideas. Lex: I mean, all walks of life, I have seen it. Rogan inspires one of my reasons, which is often celebrating for others, rather than causing sustained competition, as if the resources are very rare. When you celebrate others, when you promote the idea of others, what will happen, make big cakes, and resources are not so scarce. This applies to a lot of different areas.
It applies to academia, where many people think of academic research is zero. In fact, if you celebrate each other, if you let everyone feel excited about artificial intelligence, physics, math, I think there will be more and more funds, I think everyone will win. Yes, I think this is widely applicable.
Elon: Yes, it's okay.
3.3 About Love
Lex: About the last issue of love and meaning. Widely, what is the role of love in humans, what is it for you? Love, romantic love or other kind of love, how to make you better people? Or better engineers?
Elon: Now you have begun to ask a very confused problem. It's hard to give ...
I mean, there are a lot of books, poetry and songs, what is love, "What is love? Baby don't hurt me?"
Lex: That is one of the best answers, yes. You have quoted Shakespeare before, but this time answered is higher.
Elon: Yes, love is a very good thing.
Lex: Because we have already talked a lot of inspirational things, such as useful, solving problems, and relieve pain. But it seems that the connection between people and people is a source, it is a happy source, it is the source of the meaning, that is, the so-called love, friendship and love.
I just want to know if you consider these things, when you talk about the light of human awareness, when you talk about us to become a multi-planetary species. At least for me, this means that if we are just a person, conscious, wisdom, this is not as important as we are with others, right?
When we are together, some magic will be produced. The magic of friendship, I think the highest form is the magic of love. I think it is broad, love is not only romantic, it includes romantic love, family, and so on.
Elon: Well, I think I am concerned that we have become a multi-planetary species and space civilization because of fundamentally, I love human. So I hope to see human prosperity, do great things, and stay happiness. If I don't love humanity, I will not care about these things.
Lex: When you look at the history of the whole human history, all people who have lived, all people are still alive, we are all very good. In general, we are a very interesting person.
Elon: Yes, consider all situations. I have read a lot of history, including the most dark, worst parts. Despite this, I think that I still love humans.
3.4 Significance of life
Lex: You have a joke, 42, what do you think of this whole thing? Is there a way to represent numbers?
Elon: Oh, I should say, yes. I think Doug Sanders said in the "Galaxy Tour Guide", the universe is the answer. We really need to figure out what questions about this answer.
It is a truly difficulty part of the issue. If you can ask questions correctly, then relatively, the answer is very easy. Therefore, if you want to know about the universe, what is the problem, you want to understand the meaning of life, you need to expand the scope and size of your awareness, so you can better understand the essence of the universe and understand the meaning of life.
Lex: Finally, the most important part will be the right issue.
Lex: thus enhances the role of interviews asking people as the most important person in the room.
Elon: It is difficult to make good problems. It is absolutely true.
However, my philosophical foundation is curious about the essence of the universe. Obviously I will die, I don't know when I will die, but I will not live forever. But I want to know that we are going to understand the nature of the universe and the meaning of life, and what questions about the answer about the universe. Therefore, if we expand human and conscious range and scale, including silicon base consciousness, then fundamentally, this seems to be a good thing.
Lex: Elon, just as I said, I am very grateful, you can spend your precious time with me today.
Also, you have given millions of people in this difficult period, in this era of splitting and this cynical era. I hope that you will continue to do what you are doing. Thank you very much for talking today.
Elon: Oh, you are welcome. Thank you for your wonderful questions.
Lex: Thank you for listening to the conversation with Elon Mask. In order to support this podcast, please check our sponsor in the description.
Now, let me end with the words of Eron Mask himself.
"When something is important enough, you will do it, even if the opportunity is not good for you."
Thanks for listening, I hope to see you next time.
Guess you like
- 2022-01-23Ceiling, e-commerce is not broken
- 2022-01-23How can the Chinese tide live?
- 2022-01-22Xiaomi 10S: is about to be a thousand yuan, Xiaomi 12X: Goodbye
- 2022-01-22B station held 2021 top 100 UP main festival technology knowledge class UP main ratio doubled
- 2022-01-22Three Snapdoors flagship chips were measured, and the performance of Xiaolong 8Gen1.
- 2022-01-22How to choose the new year hollow 8 flagship?iQOO9 and Xiaomi 12 purchase suggestions
- 2022-01-22Photoshop2022mac crack version officially released!Wen dynasty address
- 2022-01-21China's eyes found 11 pulse stars, the oldest one talented 6 thousand!
- 2022-01-21Xiaomi 12PRO / One plus 10PRO / iQOO9PRO image matchup, all the new flagship, the gap is very
- 2022-01-21IqooneO5S and Realmegt2, two Snapdragon 888 models, which is more worth buying?
- Is the shoulder PM981A, comparable to the 1TBNVME solid cargo goods in SN750450?"Group of E12S Hard Disk Detoxification - Addition"
- Xiaomi 10S: is about to be a thousand yuan, Xiaomi 12X: Goodbye
- Winter Olympics scene pilot landing digital RMB will promote large-scale promotion
- Photoshop2022mac crack version officially released!Wen dynasty address
- Why do you want to be "breeding"?
- 2021 of national beauty makeup, brand is difficult to plan, traffic is difficult
- COFCO "I bought the net" was exploded to survive, and the number of goods is not as good as a street couple shop.
- Buying advice!Realmegt2Pro and millet 12pro detailed comparison
- Defeat SF, the first courier company in the word of mouth is easy to win, from the loss of 12 years to 25.7 billion
- Counting the nine series of iPhone phones, belongs to every era, reviewing the hot reasons for people