Reborn in 2010, the AI ​​in my mind is too powerful.

Yesterday, Chen Yiran was a bit tired from running Jasmine. On the new day, since she didn't have to go to class, she slept until after 10 o'clock before being woken up by a phone call.

It was a call from AVE informing them that they had received the bill of lading and sent it to their bank. They also sent them an original copy.

Chen Yiran wasn't fully awake yet, so she hummed twice and hung up the phone.

I lay in bed for another ten minutes or so before I felt like I was finally awake.

The ocean bill of lading is the most crucial document in the entire parallel vehicle transaction. It's practically a cargo receipt issued by the shipping company, proving that those thirty X5s were indeed loaded onto the ship, departing from the Port of Los Angeles, with their destination Tianjin New Port.

AVE will package the bill of lading together with the invoice, loading list, certificate of origin, and vehicle conformity certificate and hand them over to their bank. After the advising bank verifies and signs, it will send the complete set of original documents to the Guobao Bank where Chen Yiran's company is located.

Normally, Guobao Bank should review and then disburse the loan, but Chen Yiran used a shell company to open a back-to-back letter of credit.

Guobao Bank will likely just glance at the documents and then send the complete set of documents via DHL next-day delivery to Manager Wang's issuing bank in China. Anyway, no matter what problems arise with the goods, once they pay Guobao, Guobao will pay them back via AVE.

So in another four or five days, I'll have the money.

"Jasmine."

【exist. 】

"Shall we start writing the framework we discussed yesterday today?"

[Sure. Let me first explain the overall structure. We can name it DeepFlow. The core consists of two parts: an automatic differentiation engine and a network layer library.]

Speak like a human.

The automatic differentiation engine calculates derivatives for you. You define a network structure, tell it what the input and output are, and it automatically calculates the gradient for each step without you having to write the formulas yourself. The network layer library is like building blocks: convolutional layers, pooling layers, fully connected layers, activation functions—you take the layers you need from the library and build them up like Lego bricks.

"I still don't want to speak like a human, so what's next?"

After the framework is set up, you feed in the training data, and it automatically calculates the forward propagation, which means the data flows from the first layer to the last to get the result. Then it calculates the error and propagates it back layer by layer through backpropagation to update the parameters of each layer. After repeating this process thousands of times, the model is trained.

"Okay, can our small computer handle it? Don't we need to rent any extra servers or anything?"

[Not really necessary. We'll write the automatic differentiation engine this morning—that's the foundation. We'll write the network layers this afternoon. Tomorrow we'll build the convolutional network. The day after tomorrow we can run MNIST.]

"Wow, you're working that hard?"

[If we go all out, it should take four or five days.]

"Don't go all out. Take it slow, just write for a few hours each day."

Why?

"I feel uncomfortable when you run at full speed. I've already been reborn, why should I suffer so much? What's the point of racing against time these few days?"

Okay. Then we'll go at your pace. About seven or eight hours a day.

"Get lost. Seven or eight hours is like going to work. Let's say two hours a day."

Chen Yiran got up, washed her face, made a cup of coffee, and opened her terminal.

Another day of turning into a typewriter. The things Jasmine gives are getting more and more complicated, but Chen Yiran's typing speed is also getting faster and faster.

First, the automatic differentiation engine was set up, followed by numerical differentiation and caching. Jasmine said this approach was different from Theano.

Theano's method of using symbolic differentiation requires building a computational graph before compiling it into C code, and the error messages during debugging are completely incomprehensible. Her solution calculates directly in Python, which is slower, but if an error occurs, it's immediately clear which step went wrong.

Chen Yiran played for more than an hour, and then her hands started to hurt, and her head started to feel dizzy. She quickly grabbed some chocolate and ate a lot of it.

[Next, we need to work on the network layers: convolutional layers, pooling layers, fully connected layers, and the ReLU activation function.]

"We'll talk about it later. Is our automatic differential engine ready today?"

Not yet. At your pace, it should be finished in another 3 to 4 hours. I've tried my best to minimize the use of models.

"I won't do it then. I'll just do it for an hour and a half each day from now on. I'm still feeling really unwell, and I don't even have the energy to play games. Why do I feel even more dizzy and lightheaded this time than the last time I played something?"

Relatively speaking, I thought about this more this time.

"Okay, now let's clear our minds. Even large models need a rest."

I took a nap and was just about to have dinner when Manager Wang called.

"Xiao Chen, I've been notified by the Bank of China about the bill of lading. DHL has already sent it to us, and it should arrive tomorrow. Once the Bank of China reviews it and finds no problems, they'll transfer the funds to you. Usually, it takes at most two days to process the review, and the money will definitely be wired to you in a day or two more."

"Okay, whatever kind of vehicles you need, I can get them all. This isn't the first order, so you know my capabilities. There's no problem with the vehicle supply."

"I definitely need both Cayennes and GL350s, as many high-spec models as I can get my hands on. I need the diesel GL350s too. I also need the X6 xDrive50i now; if you can get your hands on one, it'll be easy to sell. The market is good lately; I have plenty of quotas available, as long as you can get the cars."

It's no surprise that the market is booming; this is precisely the time when domestic 4S dealerships are acting unethically. Many popular models are even being sold with markups of 300,000 to 400,000 yuan, and even then, there's no guarantee of having a car in stock.

Therefore, the profit margin for parallel import cars has become larger. Sometimes, they are sold at the same price as official import cars, since these are in stock and even at the listed price, they are still hundreds of thousands cheaper than those at 4S stores.

"The Cayenne is unlikely to sell well. It sells too well, even in the US. Dealers are generally reluctant to take the risk of exporting. However, I should be able to get the GL350 and X6. Should I order 30 of each model first? But let me tell you, this is all done in batches. Once these three models are done, it will be impossible to get the same models in such large quantities. The manufacturers aren't stupid."

"Okay, okay, let's focus on these popular ones first. There are also people who want Cadillacs, but the profit margin will be a bit thinner."

"Okay, I'll go ask the dealer about the situation first. I'll send you the window tinting paper first, and you can handle the letter of credit."

Chen Yiran hung up the phone and called AVE.

When the call connected, the tone on the other end was noticeably warmer than last time. The first order of thirty X5s went without a hitch; the bill of lading was submitted to the bank, and all that was left was to receive payment. This was clearly a major channel.

"Chen! Is there anything else you need to say? We can't adjust too many X5s in the short term. I can easily gather another 30 from the east and central regions next month. I've already gathered 30 GL350 diesel models."

"Okay, send over the window film first, then I'll need to calculate the price and give you a quote. Can you get 30 X6 xDrive50is?"

"The X6 50i isn't as sought-after as the Cayenne, but the supply is limited, so you'll have to give me an extra week."

"Okay, so what's the approximate price? Putting aside the paperwork, let's calculate based on the higher-end configurations of these models. What's your price?"

"GL350: 70,000; X6: 80,000."

"Okay, let's proceed this way. We'll start with 30 units of each model. I have a large sales volume here, and you can reserve any extras by letting me know."

"Okay, I'll contact you again if anything happens."

After hanging up the phone, Chen Yiran began to calculate. It was already February, and 2010 was the year with the most insane price gouging for imported luxury cars in China. The previous X5 was just a trial run; the GL350 BlueTEC diesel, which wasn't officially imported to China, and the X6, a popular model among luxury cars, were the real profit drivers.

Roughly calculated, if each vehicle is charged an extra 20,000, it seems that domestic customs clearance companies could still make a profit of tens or even hundreds of thousands of yuan.

After all, the BMW X6 priced at 180 at the 4S store requires an additional 300,000 yuan and you still have to wait several months.

That's settled then. I'm going to Stanford soon, so I need to buy a house in Silicon Valley.