Chapter 386: Musk’s Inner Monologue: Already Behaving
Musk in this world’s timeline and Musk in the timeline without Lin Ran are completely different.
In the timeline without Lin Ran, America didn’t have any urgent need to land on the Moon at all, let alone the best Moon Base site selected by NASA being occupied by China, with the entire Lunar South Pole facing the risk of being lost.
Or rather, it had already been lost.
So, after his bet succeeded, Musk had the energy to mess around with DOGE and that so-called Government Efficiency Department.
Although this department claims to do massive layoffs, to point the knife at federal employees, to give Washington a thorough big cleanup, sweeping away all the insects in Washington.
And cutting at least 2 trillion US dollars in annual federal spending, reducing federal institutions from 400 to 99, laying off 1.65 million federal employees and keeping only 550,000, introducing artificial intelligence to improve the overall operating efficiency of the government.
With quite the momentum of swallowing thousands of miles like a tiger, scaring Chinese Internet netizens badly: Is Musk playing for real? Is America coming back!
However, in reality, the biggest effect of DOGE was being established and then laying itself off.
American media claim that DOGE redefined fraud.
No benefits from layoffs were seen, but the downsides were a bit too direct.
DOGE, established on January 20, quickly laid off 400 people from the Federal Aviation Administration FAA responsible for managing federal civil aviation, followed by one plane crash after another.
The most direct one was in Washington D.C., where a passenger plane collided with an Army helicopter.
In this timeline, America faces a much more difficult situation than in the original spacetime, so Musk naturally has no ambition to establish DOGE; his goal is limited to NASA, to make use of the trillions of US dollars.
To let America achieve final victory in this space race.
DOGE, government efficiency—this has nothing to do with me; what I want is only NASA’s efficiency, and what I want to sweep away are only NASA’s insects.
Since he doesn’t want to do these things, facing Lin Ran’s accusation: Did even 1 US dollar of the 100 US dollars go to refugees?
Lin Ran’s words are turning the accusation against illegal immigrants into an accusation against Commissioner Smiths.
If it were just Commissioner Smith, that would be one thing, but more importantly, this fundamentally denies the Elephant Party’s theory that illegal immigrants are harmful, denying Big T’s definition of illegal immigrants.
The latter is the topic Musk dares not touch, or the slippery slope, where one slide easily leads to Big T himself.
Musk knows very well that Big T is definitely watching this live broadcast.
He once heard it firsthand on Big T’s private jet, where the other claimed to admire Randolph greatly, saying that back then Randolph should have been kept in America, and China shouldn’t have gotten their top-tier talent back without paying any price.
Similar admiration he has heard many times from Big T’s mouth.
Including before this trip to China, Big T also said to him: “Elon, I’m really looking forward to your dialogue with Randolph!”
Musk can even imagine the scene of Big T drinking cola in his manor watching the live television broadcast.
“No, Randolph, we need to clarify a fact, that is, what you said has no data support.” Musk said.
Lin Ran retorted: “What you said also has no data. In the absence of data, pointing the finger at illegal immigrants and directing the resentment of ordinary Americans whose lives are not going well toward the illegal immigrant group—isn’t that an even more irresponsible behavior?”
The reason Lin Ran is so fixated on this topic is that back when he was in America, he was also an illegal immigrant.
Not an illegal immigrant entering from the border, but an illegal immigrant entering from a spacetime rift.
Plus, in New York, in Washington, he was in the era when the Civil Rights Act was passed, when America was moving from racial discrimination to racial equality; he stayed in this era for nearly a decade, so Lin Ran clearly can’t stand the Elephant Party’s current blame-shifting behavior or Big T’s strategy on immigrants.
Musk hurriedly said: “Randolph, we’re not professionals; I think we shouldn’t continue this fruitless topic.”
Already sweating profusely, really sweating profusely.
If it continues, Musk feels he will keep getting beaten down.
On the Chinese Internet, in America, and in Europe, the bullet screens in live broadcast rooms in different regions are completely different.
“Watched so many of Musk’s live broadcasts, first time seeing him with such a constipated expression.”
“Feels like he was ambushed”
“No preparation at all”
“Ran Shen is too naughty, always hitting people’s weak points.”
“Ran Shen is speaking up for the immigrants; they are almost all illegal immigrants, but I bet this group of immigrants will turn around and spray Ran Shen on Twitter”
“Brainless immigrants are like this”
“So satisfying, first time seeing this kind of drama, making Musk, who just succeeded in betting on the big election and is at the peak of his life, so at a loss.”
“Musk’s inner OS: Already behaving”
“What is America-pass? At least you have to be at Ran Shen’s level to say you understand America? Those online parrots of freedom faction should learn well!”
On YouTube, whether white leftists, rednecks, or minority ethnic groups, none have anything good to say.
The former is because Lin Ran dares to question the Donkey Party pretending to build housing for illegal immigrants while actually pocketing huge sums—you dare to question the Donkey Party?
The latter is because how dare you speak up for illegal immigrants?
Minority ethnic groups actually hate illegal immigrants more than White People, with the mentality of having just gotten on the bus and wanting to weld the door shut.
Instead, in places outside America, everyone is liking and agreeing, feeling that Lin Ran voiced everyone’s thoughts:
“Feels so good to see Musk eat it”
“To help Big T get elected, he said too many words against his conscience; in front of truly smart people, his lies pop with one poke”
“Hard to imagine that past Musk would fall to this level today.”
“If Big T is a cancer, then tycoons like Musk are the chief culprits fostering the growth of that cancer.”
European netizens definitely don’t like Big T. We keep bringing in illegal immigrants by the boatload, all because of America’s mistakes in the Middle East, and Europe has to foot the bill. Not only that, but the painstakingly deployed Nord Stream 2 got blown up, sending energy prices skyrocketing.
Old Europe specializes in hating anything related to Big T, but that doesn’t mean they particularly like Lin Ran.
Lin Ran then smiled and said, “Okay, I don’t want to dwell on these things either.
The reason I want to say this is out of humanitarianism. America treats them as consumables, then for political reasons tries to direct Americans’ anger at this group.
They originally have no identity, no status, can’t enjoy America’s welfare treatment, get the lowest salary, live a precarious life, yet bear a notoriety they far shouldn’t have to bear, Elon.
Okay, I’ve said what I wanted to express. Let’s talk about the next topic.”
After Lin Ran said this, the live broadcast audience might not feel it, but the reporters in the audience and Apollo Technology employees could clearly sense that Musk breathed a sigh of relief, and the tense atmosphere in the entire venue was swept away.
“No way, how is Ran Shen’s presence even stronger than Musk’s?” Apollo Technology new employee, employee number 21101 Li Yiqing asked the team leader beside him.
He had always grumbled inwardly about this number, complaining plenty in his friends’ group: My bachelor’s, master’s, and PhD are all from Yenching University, how did I become 211 now? This downgrade is a bit too much.
The team leader said softly, “The professor gets really terrifying when he gets angry.
If Apollo Technology has any rules of strangeness, the first one is definitely: better to admit a mistake than to lie. No matter what field your work is in, don’t try to fool the professor—you definitely can’t fool him.
Even if he knows nothing about the research you’re doing and doesn’t understand it at all, his instincts in the science field will make him detect something wrong.
The next time you’re called in by the professor for a talk, you’ll find he’s become an expert who knows more than you, and if you really deceived him, depending on the severity, you’ll enjoy treatment ranging from marginalization, retraining, to winning the layoffs gift package—pick one.
Anyway, this is an iron rule at Apollo Technology.”
Li Yiqing understood: “Got it, don’t worry.” He had heard the story of Ran Shen slaying a big problem after warming wine, and how Xu Xian’s PhD big problem was slain in five minutes—he wouldn’t be so foolish as to think he was smarter.
The other reporters also felt surprised.
Presence is a very metaphysical thing.
You can’t see it, but it definitely exists.
They thought Lin Ran was coming to interview Musk with a learning mindset. After all, Lin Ran is the latecomer, the young one. In terms of achievements, he could be said to surpass Musk, but Chinese people tend to respect their seniors. It was more like a junior consulting a senior, ending the dialogue in a friendly atmosphere.
But from just the first Q&A segment, everyone could clearly feel that Lin Ran was the senior and Musk was the junior—not a junior consulting a senior, but a senior guiding a junior.
Now this feeling was even stronger.
Even in terms of presence, Lin Ran completely suppressed Musk.
After Lin Ran finished speaking and started the next segment, even the most oblivious person could feel the presence suppression.
“Elon, how do you view artificial intelligence technology? The reason Tesla doesn’t install lidar—is it due to cost considerations, or because of the so-called first principles?
There are many ways to interpret not installing lidar. It’s that Tesla would perform better in autonomous driving with lidar than without, but Tesla performs better in autonomous driving without lidar than all competitors, so you don’t install lidar for cost reasons.
Or is it first principles—that having lidar would actually weaken autonomous driving performance? Humans drive by vision; the lidar signal would instead disrupt the entire model. What is the reason?”
This was Musk’s comfort zone. From the start of the dialogue, a confident smile appeared on Musk’s face for the first time. “Um, actually, I’ve answered this question many times.
I don’t hate lidar, but it’s unnecessary.
Randolph, when you drive, do you use your eyes or lidar?
You use your eyes.
Almost all intelligent creatures on Earth navigate using vision systems.
Our brain uses two-dimensional perception, then infers three-dimensional.
The problem with lidar is that it provides information that’s too ‘clean’—actually useless.
It makes the system lazy, not truly understanding the world, just relying on the sensor’s point cloud.
That might be great in controlled environments, but in the real world, relying on lidar is a dead end.
Besides, lidar is expensive, high complexity, poor integration.
Cars are consumer products, not NASA’s laboratory.
To achieve global scale autonomous driving, the only reasonable solution is vision.
Right, pure vision.
Tesla’s advantage lies in neural networks; we train the system to truly ‘see’ the world, like humans.
You can’t rely on a crutch—lidar is a crutch. If you rely on it, you’ll never build an AI that can drive like a human.
So this isn’t just a cost issue, nor is it that we’re ‘better than others.’ It’s physical reality, first principles: vision is the strongest sensor.
Future autonomous driving systems, if they really need to be safe, widespread, and cheap, must prioritize vision.
We’re just ahead of the curve.”
When Musk got excited, he shrugged and finally summarized: “You won’t see a pigeon flying around with lidar attached, right?”
“But bats have organs that use echolocation similar to radar. China has already reduced lidar prices to 1,000 RMB, equivalent to less than 200 US Dollars. Three years ago lidar was very expensive, but the situation is completely different now.
In this case, do you still stick to your idea?
I also do artificial intelligence, and I instead feel that lidar brings new information, helping artificial intelligence better understand the entire world. You say lidar makes it lazy; not actively embracing more data is the real laziness.
In the past, we thought model training algorithms were fundamental. Now we discover that training large models with big data gives birth to sufficiently usable LLMs. Lidar ultimately enriches data volume and increases data dimensions. From the three principles of algorithm, computing power, and data, having it is definitely better than not.”
“Why not? Blindly pursuing reality replication instead of incorporating lidar information into the model—isn’t that also a form of laziness?” Lin Ran countered.
Musk was still thinking briefly.
The bullet screen on Youtube had exploded, “???”
“200 US dollar lidar?”
“Not 200 US dollars? Even less?”
A uniform row of question marks.
In America’s memory, they were still stuck with the impression that lidar cost tens of thousands of US dollars each.
Velodyne’s lidar unit price was as high as 75,000 to 80,000 US dollars, of course that was the price from a few years ago.
But even last year in 2023, the average shipment price of lidar throughout the year was around 4,000 US dollars.
Now China’s big shot comes out saying less than 200 US dollars each, why not use it?
One can imagine the shock to American netizens who know the price; the number is right there, comparable to the impact of Xiaohongshu bill reconciliation.
“Bats do use sonar, but humans are not bats.
What we want to build is cars for humans to drive, not cars for bats to drive.
Humans have eyes; the brain drives using vision.
If you want the car to learn to drive like a human, then use the human perception method.
You say lidar is cheap now, yes, much cheaper than three years ago, but the key is not the price.
The key is that it fails to solve the core problem: understanding the world.
Lidar gives you a point cloud, a distance field; it doesn’t know if the traffic light is red or green, it doesn’t know if this is a child or a plastic bag.
It’s just a more expensive ruler.
In the artificial intelligence field, more data doesn’t equal good data.
Of course we can pour more inputs into the model, but the information must have information entropy, signals useful for understanding the world.
Lidar may be useful on highways or in highly rule-based scenarios, but in daily urban driving, it will instead make the system dependent on a shortcut, not building a true world model.
You mentioned large language models, yes, scale is important.
But the premise of language models is that human language itself contains the complexity of the entire world; visual input is the same, vision contains all the complexity needed for driving.
If we add lidar, the neural network might rely on ‘simplified answers’ instead of learning the truly difficult but critical parts.
So, this is not laziness.
On the contrary, pure vision is the harder path, requiring stronger networks, more computing power, smarter training; lidar is a shortcut, but taking the shortcut won’t get you to the finish line.”
Musk spread his hands and smiled: “If you’re willing to put a bunch of sensors on your car roof and make a ‘high-priced research toy,’ of course you can.
But if you want tens of millions of cars worldwide to achieve autonomous driving, you must take the vision route; there’s only one path to a scalable future.”
Lin Ran said: “Multimodal multimodal, if even vision and sensor data can’t be integrated into the same world, then what multimodal are we talking about.
If what we really pursue is simplicity, singularity, how humans understand the world is how artificial intelligence or robots should understand the world—this is arrogance; humans shouldn’t have cars, airplanes, trains, humans should just rely on their legs, constantly honing their legs.
Today’s large models shouldn’t be trained on various structured and unstructured data; it not only dialogues but also outputs images and animations.
From data to images to animations, the outputs of large language model large models are becoming more diverse, yet autonomous driving still clings to fundamentalist artificial intelligence, believing it should only be vision.
Now if we have a framework that can incorporate lidar data into the entire model, truly achieving generalized autonomous driving, Elon, would you think you’re wrong?” Lin Ran asked.
Musk was already used to the previous points; he often saw these rebuttal viewpoints in the background, on Twitter, even internally at Tesla.
Is Tesla’s refusal to use lidar purely first principles? Or is it saving cost as netizens say?
Actually neither; at the beginning it was because of cost, lidar was ridiculously expensive at first, so expensive that a single lidar couldn’t cost tens of thousands of US dollars.
So at the start, Tesla formulated a pure vision autonomous driving proposal.
Later, having invested heavily in this technology route, the massive sunk cost made it impossible for Tesla to turn around to camera and multi-sensor fusion technology routes—who knew China could bring down the lidar price so quickly.
Who would have thought that in just three years, it would be beaten down to a true cabbage price of less than 200 US dollars.
Everyone invests real money; Huawei can’t do a pure vision proposal for the same reason.
Sunk cost is not so easy to abandon.
Lin Ran continued: “Perfect timing, rather than choose a day, hit the day—the car built jointly by Crimson Technology and Huawei, equipped with the world’s first quasi-L5 level autonomous driving technology, will park downstairs soon. You’re invited to try it and see if you can change your attitude.”
Musk murmured: “L5?”
Autonomous driving is divided into L1 to L5; L3 means the automatic system can complete some driving tasks, the driver can hand over tasks to the system but must be ready to take over at any time.
China’s mainstream is basically stuck at this stage; it’s just that due to responsibility allocation, everyone only dares to say in promotion that this is L2.99999 with infinite 9 cycle.
L4 is pure autonomous driving under specific conditions; Radish Run quick run and California’s Waymo belong to this level, full autonomous driving in specific areas.
L5 is autonomous driving without condition restrictions, the system completes all driving tasks under all conditions.
Lin Ran nodded: “Of course.”
As a top mathematics master, Lin Ran’s first thing after cooperating with Huawei was to create an autonomous driving framework under multi-dataset fusion.
Huawei has the data, Lin Ran provides the algorithm, and at the chip level, the semiconductor process optimization brought by the Moon fills in the last shortcoming.
Now their only problem is that although the chips are produced locally in China, they still rely on ASML lithography machines, and domestic lithography machines have not yet been conquered, so on this basis, your production capacity cannot go up no matter what.
Ascend computing card, Kirin chip, autonomous driving chip, all of these depend on the few ASML 7nm process lithography machines.
Production capacity is right there, it’s hard for you to ship on a large scale.
And precisely because of this, introducing the next-level Japan technology becomes so important, which is an immense benefit for China.
Musk then asked: “What about cost? How much is the hardware cost for multi-sensor?”
Lin Ran held up five fingers: “Total will not exceed 50,000, the unit is RMB.”
This is exactly 10,000 cheaper than a set of Tesla’s FSD, it’s hard to believe this isn’t intentional.
“???”
“Will it be used on Zun Jie?”
“L5?”
“Competitor mentality collapsed”
“My Mi’s good days have only lasted less than half a year, and now you pull out a big killer like L5?”
“50,000 cost? Including R&D, let’s say the cost per unit is 100,000, lowest pricing down to 300,000, aimed right at my Mi?”
“Ran Shen, have you forgotten the time you spent with Lei Zong on the Moon?”
“Musk is dumbfounded, my FSD hasn’t even entered China yet, and now you’re telling me China already has L5?”
The bullet screen exploded like a pot boiling over, with netizens’ discussions particularly heated.
This is L5.
No Classic of Mountains and Seas naming, no press conference, just quietly pulled out like this.
“Indeed Ran Shen style, last year Deep Red didn’t even hold a press conference, iterations weren’t even announced, letting everyone explore on their own, isn’t L5 this style very normal?”
“I’m really curious now how Huawei will name it when they sell it.”
“I say don’t do any Classic of Mountains and Seas naming, that really only attracts AAA building materials king Zong, can’t attract young people, better to just call it Deep Blue which is cooler”
“Still call it Ran, Honda can do Ye, Honda·Ye, why can’t Huawei and Deep Red do Ran?”
The rest of the process left Musk distracted.
Talking about moon landing, he said America will build a base on the Moon under the leadership of great President T, and successfully land on Mars.
Talking about reusable rocket, he said Starship has so many advantages, then praised Burning One Modified a couple times, just distracted.
Lin Ran finally summarized: “Elon, welcome to Shanghai, shall we go take a look now?”
Musk nodded and stood up: “Good!”
On the road to Apollo Technology park, a black electric vehicle was already parked, with clean and smooth body lines, no traditional rearview mirrors, no abrupt mushroom-head-like sensors, all sensors smoothly embedded into the body.
The identification on the side of the car reads four short letters, not AITO, but TEST.
“Full-domain L5, no regional restrictions, whether Shanghai, Yudu or WLMQ, whether heavy rain, snowy weather or heavy fog, the vehicle can independently complete driving tasks.”
After studying it for a long time, Musk asked: “What about extreme weather? Like heavy rain, thick fog, blizzards?”
Lin Ran said: “Still let it operate, in these extreme weathers, if you still need to travel, I believe it is more reliable than the vast majority of human drivers, after all it has radar, even if visibility is blocked, it still performs perfectly.”
“Technology-wise, switching from normal weather to extreme weather, how do you adjust parameters to make it rely more on radar rather than vision in such conditions?
More specifically, how do you do it to achieve balance between vision and sensors in the model?”
“In normal weather, vision is dominant because it carries the highest dimensional information.
The model first relies on end-to-end world modeling from vision.
I have never doubted this.
But when sensors detect the environment entering extreme conditions, such as raindrops interfering with camera imaging, or haze causing a sharp drop in contrast, the system automatically triggers multimodal weight adjustment.
Not manually changing parameters, but introducing cross-modal adaptive mechanisms during the training phase.
In other words, there is a dynamic perception gating unit in the model that real-time evaluates the signal-to-noise ratio and confidence of each sensor.
For example in thick fog, the confidence of the vision channel drops, the confidence weights of radar and millimeter wave automatically increase, ultimately output to the planning module in the fusion layer.
We call it weighted consensus mechanism.
Vision, lidar, millimeter wave, not isolated voting, but mutually constraining through spacetime consistency checks, once one side has hallucination or noise, other modalities immediately correct it.
In training method, we used large-scale cross-weather adversarial data augmentation.
Not just sunny, rainy, snowy days, we also simulated dust storms, strong typhoons, nighttime aurora interference and other extreme scenarios.
The model is already accustomed to modal weight transfer in the pre-training phase, it’s not learning temporarily in extreme weather, but already knows how long ago.”
While Lin Ran was speaking, the bullet screen in the live broadcast was all: “Ran Shen, enough, enough!”
“Don’t say anymore, old Ma is too sly, this is prying for intelligence.”
“Ran Shen is still too pure, if it were big mouth, he’d throw all kinds of Classic of Mountains and Seas terms at you, fooling Musk into a daze.”
“I can even imagine how Big Mouth would say it: We have a complete Tiangong System, it’s like the divine beasts in the Classic of Mountains and Seas, able to walk freely in wind, snow, thunder, and lightning.
Traditional perception is seeing, but our system is omni-perception, not just vision, radar, millimeter wave, but smashing them, melting them into a super-perception matrix, under any conditions, the car can spread its wings like Kunpeng and fly freely.
Ran Shen mentioned modal weight transfer to him!”
Lin Ran couldn’t see the bullet screen in the live broadcast, nor did he care, this wasn’t anything, the train of thought isn’t important, details are the most important.
He continued: “If using a more straightforward metaphor: vision is the eyes of the brain, radar is the sense of balance in the bones.
Normally, the eyes dominate. But in extreme situations, we let the sense of balance take over, ensuring the entire body still stands steady.
So it’s never just vision!
It’s not us artificially telling it ‘now rely on radar’, but the model itself learning in training when to trust whom more.”
Musk suddenly fell silent, he circled the car carefully, as if visually estimating how many sensors the car had and how they were distributed.
He also knew that these details needed to be captured the most.
The car door slowly opened automatically, the interior lights were soft, there was no steering wheel on the dashboard, only a slender display screen projecting a real-time 3D city occupancy map.
Musk made a joke, not to liven up the atmosphere, more like to regain his calmness: “No steering wheel, this is either a work of genius or an act of madness.”
Lin Ran and Musk got into the car, the seats automatically adjusted the angle, as if recognizing the passengers.
The moment the car door closed, the outside noise and cold were isolated, at the instant of closing the door, the reporters outside were frantically taking photos.
The photographer in the back seat continued aiming the lens at the two.
“Silly Girl, head to Lujiazui.”
Several Lujiazui locations appeared on the HUD display in front of Lin Ran.
“The first one.”
The vehicle had no jerk at all, like an invisible hand pushing it.
It slowly drove out from Apollo Technology, directly merging into Shanghai’s busy elevated road.
The closer to Lujiazui, the more complex the traffic, especially on old streets, electric vehicles, motorcycles, bicycles could appear at any time.
A delivery rider’s electric vehicle suddenly darted out from a gap, Musk instinctively wanted to raise his hand, but the car body just lightly decelerated and shifted sideways, as if it had already predicted it.
On the display screen, what Musk saw was an image precise to the millimeter: every car, every pedestrian, even the flickering rhythm of distant blurry traffic lights, all predicted in real time by the model.
No latency, naturally no hesitation.
Musk had experienced autonomous driving from countless China car brands, and also the most advanced autonomous driving technology in Tesla’s laboratory, but none were like this, so, so smooth.
Yes, smooth.
It was like a human truly driving, and an old driver at that.
Because the destination was what Lin Ran said, there was a possibility that this entire stretch of road was the result of the most data training, unable to perform so perfectly in other areas.
But it was also terrifying.
This meant that Tesla’s past massive investments in technical routes might very well be wrong.
He felt like the whole person was unwell.
Lin Ran was very relaxed, by the time the car headed toward Lujiazui, the area was full of surrounding workers, everyone skipping work to come watch Lin Ran and Musk’s arrival.
The staff sitting behind reminded: “General Manager Lin, there’s no crowd control here, don’t get out of the car.”
Lin Ran thought, what a pity, he originally wanted to have a “discussing heroes over wine” with Musk in Lujiazui, now they could only go back.
“Silly Girl, head to Apollo Technology, the first one.”
