Nearly thirty years old, established in science and technology

Chapter 20 It must be a revolutionary industrial software
Wife, you go to bed first, I still need to sort out some technical ideas, and I have some inspiration now. "After dinner, and after struggling for a long time, Ji Hong patted Cheng Hui and said softly.

"Husband, our life is very good now. Don't work as hard as before. Your health is more important!" A trace of distress flashed in Cheng Hui's eyes.

"Just for a little while, I'll sort out my thoughts. It won't be very late. You go to bed first, be good~Mu~A!"

Cheng Hui also knew that if inspiration was not sorted out immediately, it might be gone tomorrow, which would make it harder for her husband in the future, so he did not persuade her further: "Then you must go to bed early! Kiss~"

"Hmm~"

……

The AI ​​trend prediction model was recorded two days ago when Ji Hong obtained relevant design and training inspiration.

"This thing is really complicated. Without controlling the variables, even the [quasi-thinking model] is beyond its power." Ji Hong closed his eyes and silently thought:

[If you use it to predict the trend of A-share stocks, and then buy it with huge sums of money, you will definitely lose money without even thinking about it! 】

[It would probably be better if you didn’t buy it. After all, once an incident is involved, it is itself a variable, and the consequences of such a variable that affects the whole body are the most unpredictable. 】

[Also, if it is used to predict future economic trends, does it also need to add this super powerful AI capability itself to the variables? 】

[We already have this kind of super powerful AI. Will it have any impact on the economy? impossible! It itself is part of the production tool, can generate economic value, and even change the overall situation if it guides policy formulation and implementation.

[But once you use yourself as a variable for prediction calculations, the unmeasured factors will increase exponentially. This chain effect will definitely drain the model's own ability and then bring it down. 】

Ji Hong remembered again the basis of the [similar thinking model] [clustered convolutional network].

"The relationship between the whole and the part, the relationship between the individual and the cluster, the [time-influence] curve model..."

"What if the local comprehensive model of time and impact is simplified into [clusters] and are included in the prediction model as [overall variables] that contribute to predicting things?"

"Different [clusters]...weights..."

Ji Hong's mind was spinning rapidly, combined with the technical capabilities he had acquired today - the technical capabilities in the direction of GPU computing power scheduling - is this also another kind of scheduling?

These days, Ji Hong's brain is getting brighter and brighter, and his ability to draw inferences from one instance is getting stronger and stronger - whether this comprehensive model of [time and influence] is included in the variables of the prediction model, the weight and timing of the inclusion, itself has nothing to do with it. [Scheduling] has the same purpose, but it is indeed more complicated.

But if we cut through the clutter quickly, simplify the complex, and use [cluster] as the basic variable unit——

Variables are generated every moment, and the sum of the variables generated over a period of time is used as a [cluster]...

In this case, the shorter the time span, the more refined the [clusters], the more variable parameters, the higher the corresponding prediction accuracy, and the more powerful the prediction model will be. Of course, the computing power consumption will also be greater.

The longer the time span, the rougher the [clusters], the fewer variable parameters, and the smaller the computing power consumption. Correspondingly, the weaker the prediction model will be, and the accuracy will be less.

"The idea is still good. Let's go back and test it and see the effect."

Ji Hong's mind was spinning rapidly, his hands were typing on the keyboard, and he quickly recorded his ideas and plans, slowly forming a complete set of theories and technical solutions.

"The fewer variables, the more accurate it is. Predicting stock and economic estimates is almost unsatisfactory, at best unsatisfactory, but what if it is used in industry? For example, chip tape-out success rate and performance prediction?"

Ji Hong felt that it was very promising, and there was really huge market demand for this thing! In the design and manufacturing of chips, the cost of tape-out is a cost that cannot be avoided no matter what. Moreover, the more advanced the process, the higher the cost of tape-out.

A tape-out of a 14nm chip requires more than 3000 million yuan, and a 7nm chip requires at least more than million yuan.

TSMC’s latest manufacturing process, 3nm, has almost reached a tapeout cost that many top chip design companies cannot afford. According to reports, Apple paid US$3 billion for the M10 series, which is almost to billion yuan. of tape-out costs.

This cost can only be afforded by Apple, which has deep pockets. Qualcomm and MediaTek are unable to do so and can only choose second-level processes.

But tape-out is an indispensable chip design and manufacturing process - it determines whether the chip design solution is available, and is an experimental production process that verifies the rationality of the design.

What if this AI capability is used to move this experimental production process online and into an industrial software?
"This is a simulation software that is 100% realistic! No, this should be called Zhenzhen software!"

Ji Hong's expression lit up: "If it can be done, it must be a revolutionary industrial software!"

……

The idea is very clear, but the time is not yet mature and the plan needs to be refined. Moreover, this kind of ultra-large-scale industrial software, coupled with the need for more powerful AI assistance, must be supported by a complete class thinking model, then Proprietary GPU hardware is required.

So we need to wait a little longer, and these days, Ji Hong is focusing more on the company's initial operations.

If you’re not a homeowner, you don’t know how expensive oil, salt, firewood, and rice are. If you’re not a business owner, you don’t know how complicated it is.

For several days in a row, as long as he was in the company, he would be too busy to touch the ground.

It was not until Thursday that the personnel structure was basically established, the technical team was basically formed, and a front desk and recruitment specialist were recruited, freeing Ji Hong himself from recruitment and miscellaneous work, that the situation eased.

As for Zhao Donghe, it was almost the same, and he was even more busy:
"The internal servers and code servers have been configured. The external service servers are in accordance with the arrangements. The Qianxun Cloud Server is temporarily used, and the related project development work can be carried out in an orderly manner."

"What's going on with the Yuanlai Group? Is there any progress?" Ji Hong was still thinking about the news agency, so he asked.

"Chuangjie doesn't really want to let go and wants to form a three-way cooperation." Zhao Donghe said: "The previous manager came here yesterday and said it was a mess. He was playing the emotional card and still wanted to follow me. What kind of relationship could I have with him? Moreover, While you are still here, what he told me is completely and completely false."

"Three-party cooperation? What can Chuangjie do?" Ji Hong didn't look down on his old club, but he stayed there for a long time and knew their level too well:

"It's just an outsourcing project, and the amount is not very big. Just let it go. The problem can be solved. Yuanlai Group will not hold on to it. It still wants to engage in three-party cooperation. Chuangjie is so short of business now. ?"

"Yeah." Zhao Donghe nodded:

“The agency’s search promotion business has been severely impacted by short video platforms in recent years, and revenue has been less than one-third of the peak period.

"This year we launched AIGC, but the transformation was not smooth. After we left, in addition to layoffs, we also began to strictly check attendance and performance, and all benefits were suspended, which was a disguised salary reduction."

"Ignore them for now and start working on the confirmed projects first."

Ji Hong then explained: "The three customers have signed contracts and made advance payments, and the team has basically been formed. The next step is up to you."

"no problem!"

(End of this chapter)