In the conference room, after Wang Donglai finished his polite words, the focus of the meeting came.
Wudaokou naturally has certain expectations and prospects in forming such a research team.
Facing this brand-new scientific research project, not only Yao Qizhi has not had much contact with it, but also other invited experts have not had contact with it.
You must know that the domestic Internet industry is said to be innovating, but it is just transferring some mature industries from developed countries.
This is true for social networking, games, takeaways, portals, etc.
But the real core of the Internet doesn't care at all and has no intention of doing so.
As early as 2012, Google Artificial Intelligence Laboratory proposed to enable machines to learn language and grammar.
What was the domestic Internet doing at that time?
Let’s judge!
Wang Donglai, who was at the top of the conference room, stood up after bowing.
As he stood up straighter, he exuded a unique temperament.
He has the high spirits of a young man, the self-confidence of a scientific researcher, and the aura of a business tycoon.
“I don’t know if any of you here have heard of the GPT that was presented at the Bald Eagle Artificial Intelligence Academic Annual Conference in February last year. I will briefly introduce it here.”
“In 2012, Google Artificial Intelligence Laboratory proposed that it want machines to learn language and grammar. The following year , released a language model based on GPT, which learns language by encoding input. "
"In October of the same year, Google started a pilot project of GPT and made improvements to it. Learning language models when large amounts of training data are required ”
"So, our goal is actually very simple, that is, to benchmark GPT, starting with a large artificial model that is unique in China."
"The final module function to be completed in the first phase is to be able to achieve human-like performance The same communication can even be completed such as writing emails, copywriting, mathematical calculations, etc. "
When Wang Donglai came here, the professors and researchers sitting below became a little restless.
In fact, Wang Donglai’s request was too high, completely beyond everyone’s imagination.
Communicate like a human being, you can write emails and create copywriting.
Such a function is only in the first stage according to Wang Donglai.
Just listening to what Wang Donglai said, everyone quickly calculated in their minds and came to a conclusion.
That is simply impossible.
This is simply not possible with their abilities.
Although it is only their first contact with artificial intelligence, they still have some understanding of computers. At the same time, they know how powerful they are.
If they could really do what Wang Donglai said, they wouldn't be what they are now.
Immediately, a middle-aged man wearing glasses interrupted Wang Donglai's words.
“Professor Wang, you just mentioned that the module function to be completed in the first stage is to be able to write emails, copywriting, etc., but as far as I know, such technology has not been completed even in Bald Eagle. . ”
“I have also heard about Google’s GPT project. According to exchanges with some overseas friends, GPT was first proposed by Opanai’s research team. It mainly uses neural networks and machine deep learning to make machine gets Powerful natural language processing capabilities allow the model to perform tasks such as generating text, answering questions, and performing language translation. "
" Let’s not say whether it can do this, it is just natural. Language processing is already an extremely difficult subject, not to mention that artificial intelligence involves many fields such as computer science, psychology, and mathematics. The research fields include but not limited to robotics, image recognition, expert systems, etc. "
"Even the artificial model we are currently completing is not a simple subject. It is inaccurate to say that it is an artificial model. It should be called a large language model. Large language model is also called LLM. It is a deep-use model. Learning algorithms to process natural language models can identify, summarize, translate, predict and generate text and other content based on massive amounts of data.”
“So, in this way, our research team has a very big task. Heavy, it's not me who's hitting To undermine the confidence of Professor Wang and everyone else, I am actually not very optimistic about the current team."
Yao Qizhi asked before Wang Donglai spoke: "Liu Gong, what do you think? ? As long as what you say makes sense, you can work smoothly in this team. Say whatever you want!"
The middle-aged researcher known as Liu Gong glanced at Yao Qizhi, then at Wang Donglai who was silent, and then said: "It's very simple, that is to increase the number of people!"< br>
Hearing this, Wang Donglai laughed.
"Liu Gong, I didn't expect you to have such research and understanding of artificial intelligence. It seems that this project is more likely to be completed."
"As for the question Liu Gong just raised, in fact, I We have already considered it.”
“First of all, we have this time. The project is a large-scale language model, which Liu Gong must have no problem with, and large-scale language models are nothing more than using machine deep learning to complete. We can develop a language model based on recurrent neural networks to generate coherent text, etc. "
" Secondly, there is the natural language processing mentioned by Liu Gong. I also have a solution for this, which is to use a Transformer-based decoder to complete a language model under unlabeled big data as a pre-training model, and then perform sub-tasks Doing fine-tuning on the Classifiers for the same task, realizing natural language reasoning, question and answer and common sense reasoning, semantic similarity recognition, classification, etc."
"Finally, there are the fields such as algorithms, image recognition, and expert systems mentioned by Liu Gong. , I am also prepared.”
"I'm afraid Liu Gong doesn't know me well enough. He doesn't know that my first SCI paper was about algorithms. I think I'm not inferior to others in terms of algorithms."
Wang Donglai speaks clearly and speaks in a clear tone. Calm, when he said these words, there was a silent oppressive aura.
At this moment, everyone’s impression of Wang Donglai changed drastically.
Wang Donglai was so gentle and elegant just now, and now there is such a contrast.
There was not a single curse word, and there was no change in tone.
However, his words were filled with strong confidence and domineering, which made everyone involuntarily look at Wang Donglai with admiration.
Wang Donglai's performance did not surprise anyone. It is normal to achieve such an achievement at this age.
Many people focused their attention on Liu Gong, wanting to see if he would say anything else.
Unexpectedly, after receiving Wang Donglai's answer, Liu Gong's face showed a trace of thinking.
"Professor Wang, you just mentioned the development of a language model based on recurrent neural networks and natural language processing. It sounds quite feasible. Can you explain it in detail? I feel If these technologies have been solved, our research in the field of artificial intelligence will probably be at the forefront of the world.”
"Even if this is just one aspect of artificial intelligence, application products, as long as a breakthrough is made in this field, the impact will be huge, enough to cause a huge shock in society."
Yue said , Liu Gong’s eyes became brighter and his expression became more excited.
Seeing this, Wang Donglai also had a clear understanding of Liu Gong in his heart.
Pure researcher!
I interrupted myself just now, not because I had any objection to myself, but because I couldn’t see someone talking nonsense, so I interrupted. However, after he expressed some ideas, Liu Gong was able to react in such a short period of time, which proved Liu Gong's outstanding vision and profound knowledge and technology reserves.
Wang Donglai was quite satisfied that the research team had such a highly capable researcher. A smile suddenly appeared on his face and he said: "Liu Gong, artificial intelligence has a wide range of application fields and the problems involved are also very complex. Diversification."
"For now, it is enough to be able to develop a domestic general model for artificial intelligence chat!"
Wang Donglai didn't say much, let alone here. State all the key points.
As he said, some core algorithms will be completed by him. Others cannot do it first, and secondly because of the need for confidentiality.
If you are half a step ahead of the times, you are a genius; if you are one step ahead, you are a monster; but if you are a few hundred meters ahead, it is a disaster.
When Liu Gong heard what Wang Donglai said, he opened his mouth, but said nothing more. He nodded and became quiet.
Seeing this, Yao Qizhi took over the topic.
“Everyone must have seen the situation of this research project before. Our project is currently the first in the country, so we also bear the responsibility of pioneering.”
"As for this research, Professor Wang is the core, and I am not afraid of everyone laughing when I say it. I feel that Professor Wang actually knows better than me in terms of artificial intelligence and has more knowledge reserves. Just now The discussion between Professor Wang and Liu Gong proves this point.”
"What's more, this time the research project is also funded by Galaxy Technology, and the results and patents are exclusively owned by Galaxy Technology."
"Although I am the leader of the next research work, I hope you can During the research process, show your true strength, and when you encounter problems, you can listen to Professor Wang's suggestions."
Yao Qizhi picked up the microphone and said this to everyone very seriously.
As Yao Qizhi said this, the expressions of everyone present suddenly became serious.
Actually, with the previous exchange between Wang Donglai and Liu Gong, everyone has already understood.
In fact, Wang Donglai's attainments in this area are not low, and can even be said to be higher than most of the people here.
Some interested people even counted it in detail. Not only Wang Donglai’s achievements in mathematics, but also his achievements in computers are not low. The reason why Wudaokou can cooperate so readily with Galaxy Technology on this project is probably because of Wang Donglai. reason.
Having such a genius join it has only advantages and no disadvantages for the artificial model project.
Everyone can understand this.
So, after Yao Qizhi finished speaking, everyone nodded.
"Dean Yao can rest assured that Professor Wang's achievements are here, and we will naturally not turn a blind eye."
"If artificial intelligence technology can make a breakthrough, it will be of great significance to both the country and the academic community. , we will naturally not hide our clumsiness, we will definitely go all out!"
"Just right, I have it in my hand! There are no projects to be busy with, so I can put more energy into this project. "
"What a coincidence, several projects I have are at the end of their work and handed over to them. Just teach the students and get involved in this topic!”
“…”
For a moment, everyone in the conference room expressed their opinions one after another.
Wang Donglai was not too surprised by everyone's performance.
University professors actually have a lot of topics on their hands, especially those in the capital.
As for a project, most of the work is actually handed over to the students below. Doctoral students and graduate students are the main work force.
The dirty work is all theirs, and these professors only give guidance in key places.
This is not to say that these professors are deliberately squeezing, but because they have many projects at hand and are the best choice when they have no resources.
There are also considerations for training the students’ abilities.
Of course, if one person can support all kinds of people, there will naturally be unscrupulous professors who exploit students' labor force. This is inevitable.
Even Tangdu Jiaotong University has one. However, at Wang Donglai’s suggestion, Zheng Hua and Xu Songyao imposed strict restrictions and supervision on this aspect, which actually made the academic atmosphere of Tangdu Jiaotong University better. many.
But now.
These university professors and research institute researchers have put aside other projects at hand and handed them over to the graduate students and doctoral students under their leadership, focusing their main attention on artificial intelligence big data.
The reason is very simple!
Artificial intelligence big data models have greater value.
At the same time, the feasibility is also higher.
Once the project is completed, it will be a huge benefit to them.
Whether it is publishing papers or evaluating professional titles, it is helpful.
There are three main problems in the development of artificial large models, namely data, neural network architecture and hardware.
In terms of data, because it is a universal large model, the data required is very large. Fortunately, China has advantages in this area, so the problem is not too big.
As for the neural network architecture, Yao Qizhi, a big boss, and many university professors have taken action together, and finally Wang Donglai is here, so naturally it is not a big problem.
The last part is the hardware part.
Although Wudaokou’s ‘Exploration 1000’ supercomputer can be used, it is not a long-term solution.
For future development, you must have your own machine.
This requires high-performance CPUs and GPUs, which need to be imported from abroad and subject to restrictions.
Even so, I can only bite the bullet and buy it.
Currently, domestic demand simply cannot be met.
When Wang Donglai proposed a large model of artificial intelligence, he had already thought of this problem.
Buy!
This is the only option. Before the West has imposed sanctions and the country has not caught up, these things can only be bought. (End of this chapter)