Technology invades the modern world

Chapter 386 Musk's Inner Thoughts: I've Been Honest
The Musk in this timeline is completely different from the Musk in the timeline without Lin Ran.

In the timeline without Lin Ran, America has no sense of urgency to go to the moon, let alone the fact that the best lunar base site selected by NASA has been occupied by China, and the entire lunar south pole is at risk of being lost.

Or rather, it has already fallen.

Therefore, only after Musk's bet succeeded did he have the energy to pursue things like DOGE, or the so-called Department of Government Efficiency.

Although this department claims to be laying off a large number of employees, to target federal employees, and to give Washington a thorough cleaning, sweeping away the insects in the entire Washington.

What is it about cutting at least $2 trillion in annual federal spending? What is it about reducing the number of federal agencies from 400 to 99? What is it about cutting 165 million federal employees, leaving only 55? What is it about introducing artificial intelligence to improve the overall efficiency of government operations?

With an imposing and domineering presence, Musk has terrified Chinese netizens on the Simplified Chinese internet. Is he serious? Is America coming back?
However, the biggest effect of DOGE in reality was that it was established and then it was laid off.

America's media outlets claim that DOGE has redefined fraud.

The benefits of layoffs are not visible, but the downsides are rather obvious.

The DOGE, established on January 20, quickly laid off 400 people from the Federal Aviation Administration (FAA), which is responsible for managing federal civil aviation, and then a series of air crashes occurred.

The most direct example is the collision between a passenger plane and an Army helicopter in Washington, D.C.

In this timeline, America faces a much more difficult situation than in the original timeline. Musk naturally has no ambition to establish DOGE; his goal is limited to NASA, to make use of trillions of dollars.

America must ultimately win this space race.

DOGE and government efficiency are irrelevant to me. All I want is NASA's efficiency, and all I want to eliminate are NASA's parasites.

Since he had no intention of doing these things, when faced with Lin Ran's accusation, he asked, "How could one dollar out of a hundred dollars be spent on refugees?"
Lin Ran's words were turning the charges against illegal immigrants into charges against Commissioners Smith and his team.

If it were just Commissioner Smith, that would be one thing. More importantly, this fundamentally refutes the Republican Party's argument that illegal immigrants are harmful, and it also refutes Big T's definition of illegal immigrants.

The latter is the topic Musk dares not touch, or the slippery slope, which could easily slide into the hands of Musk himself.

Musk knew perfectly well that Big T would watch the live stream.

He had heard it firsthand on a private jet owned by a man named Da T, who claimed that he admired Randolph and that Randolph should have been kept in America instead of China bringing back their top talent without paying any price.

I've heard similar praise from Da T many times.

Even before he came to China this time, Big T said to him, "Elon, I'm really looking forward to your conversation with Randolph!"

Musk can even imagine Big T sitting at his ranch, drinking Coke and watching live TV.

“No, Randolph, we need to clarify one fact, which is that what you’re saying is unsupported by data,” Musk said.

Lin Ran retorted, "What you're saying also lacks data. Without data, pointing the finger at illegal immigrants and channeling the resentment of ordinary Americans who are unhappy with their lives onto the illegal immigrant group is even more irresponsible."

The reason Lin Ran is so conflicted about this topic is because he was also an illegal immigrant when he was in America.

They are not illegal immigrants who entered through the border, but illegal immigrants who entered through gaps in time and space.

In addition, he lived in New York and Washington, during the era of the Civil Rights Act and America's transition from racial discrimination to racial equality. He spent almost a decade in this era, and Lin Ran clearly disapproved of the Elephant Party's current blame-shifting behavior and Big T's immigration strategy.

Musk quickly said, "Randolph, we're not professionals, and I don't think we should continue this pointless discussion."

I'm already drenched in sweat, really drenched in sweat.

If this continues, Musk feels he will be constantly being chased and attacked.

The comments in live streams from different regions—Simplified Chinese internet, America, and Europe—are completely different.

"Having watched so many of Musk's live streams, this is the first time I've seen him with such a constipated expression."

"It feels like we were ambushed."

"Completely unprepared"

"Burning God is so mean, why does he always target people's weaknesses?"

"Ran Shen is speaking up for Runyou, who are almost all illegal immigrants, but I bet these Runyou fans will turn around and criticize Ran Shen on Twitter."

"That's what Runyou is like—brainless."

"That's awesome. This is the first time I've seen a show like this, one that can leave Musk, who just won the election and is at the peak of his life, so flustered."

Musk's inner monologue: I've finally admitted to my wrongdoing.

"What is America? You need to be at least as knowledgeable as the God of Burning to say you understand America, right? Those liberals online who just parrot what others say should take a good look!"

On YouTube, whether you're a liberal, a redneck, or a minority, there's nothing nice to say.

The former is because Lin Ran actually dared to question whether the Donald Party was using the guise of building housing for illegal immigrants to make a lot of money. How dare you question the Donald Party?
The latter is because how dare you speak up for illegal immigrants?

Minority groups actually dislike illegal immigrants more than white people, and many of them have the mentality of wanting to weld the car doors shut as soon as they get on board.

On the contrary, apart from America, everyone gave her a thumbs up, feeling that Lin Ran had expressed everyone's feelings:

It feels so good to see Musk get his comeuppance.

"To help Big T get on stage, he said too many things against his conscience. In front of truly intelligent people, his lies were easily exposed."

"It's hard to imagine that the Musk of the past would have fallen to this state today."

"If Big T is a cancer, then billionaires like Musk are the culprits that help the cancer grow."

European netizens definitely don't like America. We're getting a flood of illegal immigrants, ship after ship, all because of America's mistakes in the Middle East. And now Europe is paying the price. Not only is Europe paying the price, but the Nord Stream 2 pipeline, which we finally managed to deploy, was also destroyed, sending energy prices sky-high.

Old Europe's general disapproval of anything related to Big T doesn't mean they particularly like Lin Ran.

Lin Ran immediately smiled and said, "Okay, I don't want to dwell on these things either."

I'm saying this out of humanitarian concern. America treats them like disposable resources, and then tries to direct America's anger at this group for political reasons.

They had no identity, no status, no access to America's benefits, received minimum wage, and lived a precarious existence, yet they bore far more infamy than they deserved, Elon.

Okay, I've said what I wanted to say. Let's move on to the next topic.

After Lin Ran finished speaking, the viewers watching the live stream might not have noticed, but the reporters and Apollo Technology employees in the audience could clearly feel that Musk breathed a sigh of relief, and the tense atmosphere in the entire room was swept away.

"No, how come the aura of the Burning God is even stronger than Musk's?" Li Yiqing, a new employee of Apollo Technology, employee number 21101, asked her team leader.

He had always harbored resentment towards this number, and had complained about it frequently in his friends' group chat: "I clearly have bachelor's, master's, and doctoral degrees from Yenching University, how come I'm now a 211 university? This downgrade is a bit too much."

The team leader said softly, "The professor can be terrifying when he gets angry."

If Apollo Technology has any strange rules, the first one must be: it's better to take the blame than to lie. No matter what field you work in, don't try to fool the professor; you can't fool him.

Even if he knows nothing about your research and doesn't understand it at all, his scientific instincts will make him sense that something is wrong.

The next time the professor calls you in for a talk, you'll find he's become a better expert than you. And if you really did deceive him, depending on the severity, you'll get one of these treatments: marginalization, retraining, or a big layoff package.

In short, this was an ironclad rule at Apollo Technology.

Li Yiqing understood: "I understand, don't worry." He had heard the stories of how the Burning God was beheaded with warm wine and Dr. Xu Xian was beheaded in just five minutes. He wouldn't be so clever as to think he could be.

Other reporters were equally surprised.

Aura is a very mysterious thing.

It cannot be seen, but it certainly exists.

They assumed that Lin Ran was interviewing Musk with a learning attitude. After all, Lin Ran was a newcomer and a young man, and his achievements were arguably greater than Musk's. However, Chinese people tend to respect their seniors, and it was more like a junior asking a senior for advice, ending the conversation in a friendly atmosphere.

As a result, in the very first Q&A session, everyone could clearly see that Lin Ran was the senior and Musk was the junior. It wasn't the junior asking the senior for advice, but rather the senior instructing the junior.

This feeling is even stronger now.

Even in terms of aura, Lin Ran completely overwhelmed Musk.

After Lin Ran finished speaking and started discussing the next segment, even the most oblivious person could feel the oppressive aura emanating from him.

"Elon, what are your thoughts on artificial intelligence technology? Is Tesla's decision not to install LiDAR due to cost considerations, or is it based on so-called first principles?"
There are many explanations for not installing LiDAR. One is that Tesla performs better in Autopilot with LiDAR than without it. Another is that Tesla's Autopilot is better than all its competitors' without LiDAR, so you're not installing LiDAR for cost reasons.

Or, to put it another way, first principles theory suggests that having LiDAR might actually weaken the performance of autonomous driving. Humans drive based on vision, and the presence of LiDAR signals could actually cause the entire model to become chaotic. Why is that?

This was Musk's comfort zone. From the start of the conversation, a confident smile appeared on Musk's face for the first time. "Well, I've actually answered this question many times before."

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 rely on their visual systems for navigation.

Our brains use two-dimensional perception to infer three-dimensionality.

The problem with lidar is that the information it provides is too 'clean' and therefore useless.

It makes the system lazy, preventing it from truly understanding the world and relying solely on point clouds from sensors.

This might be great in a controlled environment, but in the real world, it's a dead end when LiDAR is used.

Furthermore, lidar is expensive, complex, and poorly integrated.

A car is a consumer product, not a NASA laboratory.

The only reasonable solution to achieve large-scale autonomous driving on a global scale is vision.

That's right, purely visual.

Tesla's advantage lies in neural networks; we train the system to truly 'see' the world, just like humans.

You can't rely on a crutch, and LiDAR is that crutch. If you rely on it, you'll never create an AI that can drive like a human.

So this isn't simply a matter of cost, nor is it about us being 'better than others,' but rather a matter of physical reality, of first principles, and of vision being the most powerful sensor.

If future autonomous driving systems are to be truly safe, widespread, and affordable, vision will inevitably be the priority.

We're just going ahead.

Musk shrugged excitedly, concluding, "You don't see a pigeon flying around with a lidar mounted on it, right?"

"But bats have an echolocation organ similar to radar. China has reduced the price of lidar to 1,000 RMB, which is less than 200 USD. LiDAR was very expensive three years ago, but the situation is completely different now."

In this situation, would you still stick to your own ideas?
I also work in artificial intelligence, and I actually think that LiDAR brings new information, allowing artificial intelligence to better understand the world. You say that LiDAR is lazy, but it is not actively embracing more data that is lazy.

In the past, we thought that model training algorithms were fundamental. Now we have found that big data trains large models, and only large models can give birth to sufficiently useful LLMs. LiDAR ultimately enriches the amount of data and increases the data dimension. In terms of algorithm computing power and the three principles of data, having it is better than not having it.

"Why not? Isn't it a form of laziness to simply pursue realistic replication instead of incorporating LiDAR information into the model?" Lin Ran countered.

Musk is still briefly considering.

The comments section on YouTube is exploding with "???"

"A $200 lidar?"

"No, $200? Less than that?"

A sea of ​​question marks.

In America's memory, they still remember that lidar units cost tens of thousands of dollars each.

Velodyne's LiDAR units cost between $75,000 and $80,000 each, though that was a few years ago.

Even in 2023, the average price of LiDAR shipments was around $4000.

Now a bigwig from China is saying, "It's less than two hundred dollars apiece, why not use it?"

It's easy to imagine the impact on American netizens who knew the prices; the numbers are there, comparable to the impact of reconciliation on Xiaohongshu (Little Red Book).

"Bats do use sonar, but humans are not bats."

We're building cars for humans, not for bats.

Humans have eyes, and the brain uses vision to drive.

If you want a car to learn to drive like a human, then use human perception.

You said LiDAR is cheaper now, and yes, it's much cheaper than it was three years ago, but the key issue isn't the price.

The key issue is that it failed to solve the core problem: understanding the world.

LiDAR gives you a point cloud, a range field. It doesn't know if a traffic light is red or green, or if something is a child or a plastic bag.

It's just a more expensive ruler.

In the field of artificial intelligence, more data does not equal better data.

We can certainly feed more inputs into the model, but the information must have information entropy, and there must be signals that are useful for understanding the world.

LiDAR might be useful on highways or in highly structured environments, but in everyday city driving, it can actually make the system rely on a shortcut instead of building a true model of the world. You mentioned large language models; yes, scale is important.

However, the premise of language models is that human language itself contains the complexity of the entire world, and the same is true for visual input, which contains all the complexity required for driving.

If we add LiDAR, the neural network might rely on 'simplified answers' instead of learning the truly difficult but crucial parts.

So this isn't laziness.

On the contrary, pure vision is a more difficult route, requiring stronger networks, greater computing power, and smarter training. LiDAR is a shortcut, but you can't reach the destination by taking shortcuts.

Musk shrugged and laughed, "If you're willing to put a bunch of sensors on the roof of your car and build a 'high-priced toy for scientific research,' of course you can."

But if you want tens of millions of cars worldwide to be self-driving, you must take the vision-based approach; there's only one path to a large-scale future.

Lin Ran said, "Multimodal, multimodal, if even visual and sensor data cannot be integrated into the same world, then what is the point of talking about multimodal?"

If we truly pursue simplicity and uniformity—that artificial intelligence or robots should understand the world the way humans do—that is arrogance. Humans shouldn't have cars, airplanes, or trains; they should rely solely on their two legs, constantly honing their abilities.

Modern large-scale models should not be trained with various structured and unstructured data; they can not only engage in dialogue but also output visuals and animations.

From data to images to animations, the output of LLM's large models is becoming increasingly diverse, while autonomous driving still clings to fundamentalist artificial intelligence, believing that it should only have vision.

"Now, if we have a framework that can integrate LiDAR data into the entire model, truly achieving generalized autonomous driving, Elon, would you think you're wrong?" Lin Ran asked.

Musk is already used to these opposing views; he sees them frequently in the background, on Twitter, and even within Tesla.

Is Tesla's decision to refuse to use LiDAR purely based on first principles, or is it, as some netizens suggest, a cost-saving measure?

Actually, neither is the case. Initially, it was due to cost. At the beginning, LiDAR was extremely expensive, ridiculously expensive, so you couldn't expect a single LiDAR to cost tens of thousands of dollars.

Therefore, Tesla initially developed a vision-based autonomous driving solution.

Later, you had already invested heavily in this technology, and the huge sunk costs left Tesla with no choice but to switch to a technology route involving cameras and multi-sensor fusion. Who knew that China would be able to drive down the price of LiDAR so quickly?

I never imagined that just three years later, you would have reduced it to a truly bargain price of less than two hundred dollars.

Everyone is investing real money, and Huawei can't simply offer a purely visual solution either.

Sunk costs are not so easy to give up.

Lin Ran continued, "Perfect timing! There's no time like the present. The car that DeepRed Technology and Huawei have jointly developed, equipped with the world's first near-Level 5 autonomous driving technology, will be parked downstairs later. We'll invite you to try it out and see if you can change your mind."

Musk murmured, "L5?"

Autonomous driving is divided into L1 to L5. L3 means that the autonomous system can complete some driving tasks, and the driver can hand over the tasks to the system, but must be prepared to take over at any time.

The mainstream in China is basically still at this stage. It's just that because of the division of power and responsibility, everyone only dares to say when promoting that this is L2.99999 infinite 9 loop.

L4 refers to fully autonomous driving under specific conditions. Carrot Express and Waymo in California belong to this level, which is fully autonomous driving within a specific area.

L5 refers to unrestricted autonomous driving, where the system completes all driving tasks under all conditions.

Lin Ran nodded: "Of course."

As a top mathematics master, the first thing Lin Ran did after reaching a cooperation agreement with Huawei was to develop an autonomous driving framework based on the fusion of multiple datasets.

Huawei has the data, Lin Ran provides the algorithms, and at the chip level, the semiconductor process optimization brought by the moon has filled the last gap.

Their only problem now is that although the chips are produced locally in China, they still use ASML's lithography machines. Domestic lithography machines have not yet been developed, so their production capacity can't be increased no matter what.

Ascend computing cards, Kirin chips, and autonomous driving chips all rely on the very few ASML 7nm process lithography machines.

With production capacity fixed, it's difficult to ship on a large scale.

This is precisely why introducing a second-tier Japanese technology is so important, as it represents immense benefits for China.

Musk then asked, "What about the cost? What is the hardware cost of multiple sensors?"

Lin Ran held up five fingers: "It won't exceed 5 in total, in RMB."

This is exactly $10,000 cheaper than a Tesla FSD system, and it's hard to believe this wasn't intentional.

"???"

"Will we need to use it in the Venerable Realm?"

"L5?"

"Our competitors are having a mental breakdown."

"We've only had a good life for less than half a year, and you're already pulling out a killer weapon like the L5?"

"5 in cost? Including R&D, I estimate the cost of one unit to be 10. The lowest price could drop to 30, which would be right up to my budget?"

"Burning God, have you already forgotten the time you spent on the moon with Boss Lei?"

"Musk must be dumbfounded. My FSD hasn't even entered China yet, and now he's telling me that China already has L5?"

The comments section exploded, with netizens engaging in extremely heated discussions.

This is L5.

There was no naming of it after the Classic of Mountains and Seas, no press conference, and it was just quietly unearthed.

"It's definitely in the style of the Burning God. Last year, they didn't even hold a press conference for the Crimson Red, and they didn't even talk about the iteration, leaving everyone to explore it on their own. Isn't this style of the L5 quite normal?"

"I'm really curious about what Huawei will name the products when they're sold."

"If you ask me, let's not bother with naming it after the Classic of Mountains and Seas. That kind of naming will only attract Mr. Wang from AAA Building Materials, not young people. It would be much cooler to just call it Deep Blue."

"Let's still call it 'Ran' (燃, meaning 'burning' or 'vibrant'). Honda can come up with 'Ye' (烨, meaning 'bright'), so why can't Huawei and Crimson (深红, meaning 'deep red')?

Musk was distracted throughout the rest of the process.

Speaking of the moon landing, he said that America, under the leadership of the great President T, would build a base on the moon and successfully land on Mars.

When discussing reusable rockets, he mentioned the advantages of having more Starships, then gave a brief praise of the Burning One Mod, revealing a somewhat inattentive attitude.

Lin Ran concluded, "Elon, welcome to Shenhai. Shall we go take a look now?"

Musk nodded and stood up, saying, "Okay!"

A black electric car was parked on the road in the Apollo Technology Park. Its body lines were simple and smooth. There were no traditional rearview mirrors or protruding mushroom-shaped sensors. All sensors were smoothly embedded in the body.

The sign on the side of the car reads a simple four letters, not AITO, but TEST.

"Full-range L5, without regional restrictions, whether it is Shanghai-Haikou, Chongqing-Dushanzi or WLMQ, whether it is heavy rain, snow or fog, the vehicle can complete the driving task independently."

Musk examined the situation for a long time before asking, "What about extreme weather? Like torrential rain, dense fog, or blizzards?"

Lin Ran said, "It should still be operated by it. In these extreme weather conditions, if you still need to travel, I believe it is more reliable than most human drivers. After all, it has radar, so even if the view is obstructed, it is still perfect."

"Technology, when switching from regular weather to extreme weather, how do you adjust the parameters to make it rely more on radar than vision in such situations?"
To be more specific, how do you achieve a balance between vision and sensor performance in the model?

"Under normal weather conditions, vision is dominant because it carries the highest level of information."

The model primarily relies on end-to-end visual world modeling.

I have never doubted this.

However, when the sensor detects that the environment has entered extreme conditions, such as raindrops interfering with camera imaging, or fog and haze causing a sharp drop in contrast, the system will automatically trigger multimodal weight adjustment.

Instead of manually changing the parameters, a cross-modal adaptive mechanism is introduced during the training phase.

In other words, the model has a dynamic perception gating unit that evaluates the signal-to-noise ratio and confidence level of each sensor in real time.

For example, in dense fog, the confidence level of the visual channel decreases, while the confidence weights of radar and millimeter waves are automatically increased, and finally output to the planning module at the fusion layer.

We call it a weighted consensus mechanism.

Vision, lidar, and millimeter waves do not vote in isolation, but rather constrain each other through spatiotemporal consistency checks. If one of them exhibits illusions or noise, the other modalities will immediately correct it.

In terms of training methods, we used large-scale cross-weather adversarial data augmentation.

Not only sunny days, rainy days, and snowy days, but we also simulated extreme scenarios such as sandstorms, strong typhoons, and nighttime aurora interference.

The model is already accustomed to the weight transfers between modalities during the pre-training phase; it didn't learn this on the fly in extreme weather, but rather it already knew it.

While Lin Ran was speaking, the chat in the live stream was filled with comments like, "Ran Shen, that's enough, that's enough!"

"Stop talking, Old Ma is too cunning, he's trying to gather intelligence."

"Burning God is still too naive. If it were Big Mouth, he would definitely use all sorts of classical Chinese terms to fool Musk."

"I can already imagine what Big Mouth would say: We have a whole set of Heavenly Crafting System, which is like a mythical beast in the Classic of Mountains and Seas, and can move freely in wind, snow and thunder."

Traditional perception is about seeing, but our system is about synesthesia. It's not just about vision, radar, and millimeter waves, but about breaking them down and fusing them into a super-sensory matrix. Under any conditions, the car can take flight like a roc spreading its wings.

"What did the God of Burning ever mention about modal weight transfer to him?!"

Lin Ran couldn't see the comments in the live stream, and he didn't care. It didn't matter; the strategy wasn't important, the details were.

He continued, "To put it more bluntly: vision is the eyes of the brain, and radar is the sense of balance in the bones."

Normally, the eyes are in control. But in extreme situations, we let our sense of balance take over to ensure that the whole body remains stable.

So it's never just about vision!

It's not that we artificially tell it 'rely on radar now,' but rather that the model learns during training when to trust which system more.

Musk fell silent for a moment. He walked around the car carefully, as if visually estimating how many sensors were on the car and how they were distributed.

He also knew that these details were the most important to capture.

The car door slowly opened automatically, the interior lights were soft, and there was no steering wheel on the dashboard, only a thin display screen that projected a real-time 3D city occupancy map.

Musk made a joke, not so much to lighten the mood, but more to calm himself down: "No steering wheel, that's either a stroke of genius or a crazy thing to do."

Lin Ran and Musk got into the car, and the seats automatically adjusted their angles, as if recognizing the passengers.

The moment the car door closed, the noise and cold outside were shut out. At the instant the door closed, reporters outside were frantically taking pictures.

The photographer sitting in the back row continued to point his lens at the two of them.

"Silly girl, head to Lujiazui."

The HUD in front of Lin Ran showed several locations in Lujiazui.

"First."

The vehicle moved without the slightest hesitation, as if propelled by an invisible hand.

The car slowly drove out of Apollo Technology and merged directly into the busy elevated road in Shenhai.

The closer you get to Lujiazui, the more complex the traffic situation becomes, especially on the older streets where electric bikes, motorcycles, and bicycles can appear at any time.

The deliveryman's electric scooter suddenly emerged from the gap, and Musk instinctively reached out to raise his hand, but the scooter simply slowed down and shifted to the side, as if it had been anticipated.

On the display screen, Musk saw an image detailed to the millimeter: every car, every pedestrian, and even the blurry rhythm of the flashing traffic lights in the distance were predicted in real time by the model.

There was no delay, and naturally, no hesitation.

Musk has experienced the self-driving technology of countless Chinese car brands and the most advanced self-driving technology in Tesla's labs, but none of them are as smooth as this.

Yes, it's silky smooth.

It's like a real human driving a car, and a seasoned one at that.

Since the destination was mentioned by Lin Ran, it's possible that this entire route represents the result of the most data used in training, and it couldn't perform so perfectly in other areas.

But it's also terrifying.

This means that Tesla's past investment in technology may very well have been misguided.

I feel terrible.

Lin Ran was very relaxed. By the time the car drove towards Lujiazui, the area was full of migrant workers from the surrounding area. Everyone had skipped work to come and watch Lin Ran and Musk's arrival.

The staff member sitting behind him reminded him, "Mr. Lin, there's no security checkpoint here, don't get out of the car."

Lin Ran thought to himself, "What a pity. I was hoping to have a lively discussion with Musk over wine in Lujiazui, but now I have to go back."

"Silly girl, head to Apollo Technology, first one."

(End of this chapter)