Nearly thirty years old, established in science and technology

Chapter 4 I like to listen to you

"Inspiration for AI fusion operating system architecture?" Ji Hong was startled: "Microsoft's Windows 12 is said to be in this direction, including the integration of software and hardware scheduling, kernel, drivers, and applications with the intelligent capabilities of large models. This is the next generation Standard configuration of the operating system."

At this stage, whether it is a computer system or a mobile phone system, large models exist more as an additional capability. Most of them provide a software or perform shallow integration at the application layer, which has absolutely nothing to do with integration.

In this case, the operating system itself cannot use the intelligent capabilities of the large model to perform various thread-level and hardware-level optimizations - it does not support it at the kernel level.

This is also determined by the times - even the youngest Hongmeng system has been designed and planned for several years, let alone the large model. Artificial intelligence itself was still jokingly called artificial retardation at that time. Of course there would be no such thing. The underlying integration is planned.

However, what Ji Hong got at this time was only architectural inspiration, and he was still far away from being able to implement it. If he really wants to do this, at least the AI ​​capabilities and the capabilities of the operating system itself must be up to standard, neither of which he currently has.

However, working on open source Linux and starting as an AI assistant is indeed a good entry point, and my wife needs it!

"Honey, do you know what the configuration of your company's computers is?"

In the case of AI personal assistants, the model is trained in the cloud, but the deployment must be done on the device for testing, otherwise the promotion will be restricted - commercial confidentiality is a very important factor.

In this case, the computer must have an independent graphics card or NPU, and at least a certain amount of AI computing power, which is equivalent to a large end-test model.

"I don't know much about the configuration, but they sent us a list before buying it. I'll look for it and send it to you."

"This is interesting, Moore Technology's graphics card." Looking at the configuration list sent by Cheng Hui, Ji Hong also sighed.

Having said that, the scale of the Moore Technology MTT S80 graphics card is actually not small, and it is even better than my current Nvidia RTX3060.

It's just that Moore Technology's graphics card is not perfect for playing games, the game driver is not perfect, and the AI ​​ecosystem is not perfect either. As a result, it has scale but cannot perform as it should, making it useless.

But this is not a problem here in Ji Hong, scale is enough.

The class thinking micro model itself is a brand new thing, and no one has an ecology, so it doesn’t matter if there is an existing ecology.

"Purchase one according to the order, and use it for testing later!" Ji Hong immediately placed an order for the complete set at Goudong. However, since the configuration was not too mainstream, he could not enjoy the next-day delivery benefit, and it would take a few days to adjust the goods.

……

For two days, Ji Hong was busy working. When he didn't need to think deeply, he called Cheng Hui to introduce relevant technical knowledge and concepts to her. Not to mention, this was really a rare experience.

"Honey, do you understand what I said?" Ji Hong asked in a timely manner.

"I don't know much about technology, but the examples you gave are easy to understand. You can definitely understand it. And I like listening to you. I like it even if I don't understand. This makes me feel that I am getting closer to you. You didn't follow me before." I told you.”

"Well, I'll tell you everything from now on." Ji Hong felt a little guilty when he said this. At first, he really told Cheng Hui this with a +1 idea.

But the more I talk, the more I realize that communication is really a bridge for emotional warming. Sometimes, just saying "you don't understand" is just a dismissal, which is very hurtful.

……

"These are all the inspirations and ideas that suddenly burst out in me today. Now I will record them. The data of this quasi-thinking micro-model need to be sorted out and I need to apply for an invention patent." When it comes to patents, Ji Hong also specifically reviewed the documents he had with him at that time. contract signed by the company.

Because it was a labor contract that was transferred during the merger and acquisition, in order to quickly resolve the matter, the company did not get too entangled and continued execution according to the original company contract by changing the subject.

The original company did not sign any competition agreement, but later signed a supplementary agreement on copyright and patent ownership. The core terms are:
The technical achievements generated by using company resources or completing work needs belong to the company, and the company will provide corresponding rewards and subsidies.

Therefore, in addition to work, Ji Hong makes these things on his own computer at home, which are completely his own.

"That's no problem!"

After confirming that there was no problem with his contract, Ji Hong began to sort out the patent application materials - ideas and algorithms cannot be patented and need to be packaged into functional modules and products.

Moreover, the essence of patent application is to exchange publicity for protection. If it is determined that no one else can do it in a short time, many companies will choose not to apply for a patent, just to avoid others providing ideas.

Even if you want to apply, you need to control the extent of disclosure - you can't be stupid and even disclose the details.

After thinking for a long time, Ji Hong decided to apply for an invention patent - but he had to wait a little longer.

It is easy to say that it is just a thinking-like micro-model. The key is the whole set of things behind it, which also has many extensions. In the future, it will definitely involve the design and manufacturing of proprietary inference computing cards.

In this case, the technology involved in this entire route must be considered first, which is the opportunity. But the current Ji Hong obviously does not have the ability to occupy the entire line. In order to avoid providing ideas to others, he has to slow down first.

In this case, the confidentiality of the class thinking model and related training methods needs to be strengthened. Once leaked, the losses will be even greater.

However, the most critical thing is in Ji Hong's own mind. When the class thinking model was initially applied in "Cyber ​​Girlfriend", it was safe to assume that the basic core would be local.

……

On Sunday evening, the first version of the model was officially formed. Something like a toy like "Cyber ​​Girlfriend" does not need to be too rigorous - it is meant to be fun, and if it is wrong, it is considered fun.

Based on this, Ji Hong directly deployed the first version of the model and applied it to the small program in the production environment, preparing to operate, train and test it.

"This training process is not like a child learning a foreign language, but more like a young student learning a foreign language. This is much faster!

"What can children remember when they are learning to speak? They may not be able to remember what their parents call them in three days. But with the combination of large models and quasi-thinking, the memory can be said to be unlimited. It just needs to be corrected."

Looking at the training results, Ji Hong had some new understandings of the quasi-thinking micro-model in his mind. He also thought about and summarized it, and constantly improved the theory and data of the quasi-thinking micro-model.

……

"Honey, go to sleep. I have to go to work tomorrow, Monday!" At ten o'clock, Cheng Hui called Ji Hong to go to bed on time.

"Yeah, right away."

(End of this chapter)