Algorithm system

Difficulties in data collectionIt was already quite hot walking on the streets in Zijin City in early August, so Lin Yuan chose to get up early.

He didn’t choose to take the subway but rode his electric scooter instead. Since building data pathways for the system would definitely require him to visit every merchant in the area, he didn’t want to rely on his own two feet.

The area Liu Huzi covered included the most prosperous part of the entire HX District. There were countless office buildings here, and the Good Group Waimai Zijin Branch was also nearby.

Lin Yuan applied to the Hao Tuan Company for an opportunity to conduct an external delivery test because some updates to the delivery platform still needed on-site, dedicated testing before going live. Generally, programmers wouldn’t be sent out to do this themselves, but Lin Yuan, in order to go out during work hours, volunteered to take on this task.

Lin Yuan wasn’t sure what the process should be for building data pathways for merchants.

For this reason, he deliberately skipped breakfast and then found a breakfast shop early in the morning, sitting inside to observe carefully.

According to previous experience, the computing power system would provide a prompt when data loading was complete. However, after Lin Yuan sat down in this breakfast shop, he had almost finished two steamed buns, but the system still had no reaction.

Ultimately, the system didn’t give any response until he finished his soy milk and walked out the door.

“Can you only load already collected data? Can’t you actively organize and acquire it based on location?”

“As a system, you should at least have some class.”

Although Lin Yuan kept complaining and grumbling to the system, the system would not respond at all when playing dead.

On the main road in the early morning, Lin Yuan just stood by the roadside, watching the passing traffic, and fell into a slight sense of loss.

If he couldn’t complete data pathway construction for merchants, then naturally, he couldn’t complete the same construction for delivery riderss. In that case, the two most critical points in the entire delivery path and delivery problem—merchants and Riders—would be completely disconnected from the AI algorithm.

What AI, what artificial intelligence, what ChatGPT. No matter how loud its name is shouted or how high its class is hyped. In the end, it all comes down to one thing—data-driven.

Even the most powerful AI model is data-driven; data is the source of everything. Even for the computing power system, it is the same.

Data represents direction and destination; without it, even a luxury car worth tens of millions would not know where to go.

If the method of data pathway construction Lin Yuan envisioned was unfeasible, then the trouble would not only be limited to the current delivery AI algorithm optimization project, but a greater problem would come from how the computing power system is used.

The computing power system can actively complete data collection with simple guidance, compared to collecting data first and then throwing it to the computing power system. This is like the difference between autonomous driving and manual driving.

The difference here is huge.

Just as you cannot be distracted by other things when driving manually, if data must be manually collected and then given to the computing power system, then Lin Yuan will spend a lot of time dealing with such data collection issues in the future.

And a further problem is. If we take autonomous driving and manual driving as an example, if the purpose of the car is to transport people to a certain place, then the only difference is whether the people in the car can be distracted. But what if the purpose itself is not to transport people but to make the car drive to a place?

That is to say, if the purpose of driving is to make the car go from one place to another. Then autonomous driving and manual driving will be worlds apart.

Because with autonomous driving, people don’t need to be in the car. People only need to set the destination for the car and then ignore it, and one person can handle thousands of cars. But manual driving is not possible; one person can only handle one car.

What is this called?

This is called underlying principles affecting upper-layer applications.

The advantages of underlying principles often produce exponential differences when reflected in upper-layer applications.

The principle of data collection is like this.

Manual collection is like manual driving; one person can only handle one data node. But with automatic collection, one person can handle N data nodes.

If Lin Yuan really had to manually collect data, then the delivery AI algorithm optimization project wouldn’t need to be done. Because there’s no way he could sit at the entrance of all merchants every day, and sit on the back seat of all delivery riderss’ electric scooters, constantly recording the data they generate.

This is how scientific principles work. When your gaze only sees one car, you don’t feel much difference between autonomous driving and manual driving. But when you project your vision to cover countless cars, the huge gap becomes apparent.

This is also one of the reasons why so many big tech companies are willing to burn huge sums of money to bet on autonomous driving.

But that’s beside the point. Lin Yuan stood in the morning breeze at this time. The gradually rising temperature in the air was like his slowly anxious mood.

After truly embarking on the IT path, Lin Yuan gradually had two greatest insights.

One is that when encountering problems, one must habitually investigate and grasp the essence of the problem. The second is truly understanding the importance of direction.

These two points are not empty words.

Lin Yuan did not despair because of his disappointment; he began to carefully analyze the characteristics of the system’s data pathways, attempting to grasp the essence of the problem.

The computing power system can easily acquire already collected delivery data exported by the Hao Tuan Company from the backend, and it is insensitive to the total size of the data; even very large data can be loaded quickly. That is to say: the system is more concerned with the form of the data.

The collected delivery data is not yet in the vector form that can ultimately be executed by an AI model.

Delivery data is generally like this: In a certain year and month, Zhang San received an order (ID: order123) at location A, then went to location B where the merchant was, spent a certain amount of time waiting for the meal, and then followed a certain path to deliver it to the customer’s location C at a certain time.

Such data cannot be directly given to a real-world AI model for calculation. The AI refers to the thing that the AI ultimately produces, not that the thing that produces the AI is itself an AI.

This point is very contrary to common sense for ordinary people—AI is actually an AI algorithm, and an AI algorithm is produced and manufactured, but the process of this production and manufacturing is not AI at all.

This is like pouring manure on fruits and vegetables in the field, and then delicious fruits and vegetables can grow. The fruits and vegetables are delicious, but the stuff poured on them is obviously not edible.

However, this is only for real-world AI models. The computing power system is not like this; the computing power system can directly load these unprocessed data for calculation. It seems the system itself can perform data preprocessing.

Before calculation, the usual practice for real-world AI models is to process this delivery data into matrix vectors.

AI models are cold; they don’t care what the data you give them means; in their eyes, it’s all matrix vectors. So, delivery data needs to be first converted into cold numbers like [-1, 23, 321, ...].

These numbers represent real delivery data. For example, if the weather during the delivery of a certain order was sunny, a certain parameter in the matrix vector might be represented by the number “1,” and then by the number “0” for cloudy.

But the computing power system is different; Lin Yuan had tested it before. Delivery data doesn’t need to be preprocessed at all; the system can process it even if it’s loaded directly. It seems the system itself can perform data preprocessing.

This is consistent with the system’s nature—after all, this system is like a living computer that can change its hardware parameters as needed.

So Lin Yuan naturally thought of finding a breakthrough from this point.