It is normal for this kind of negotiation to not be able to go through the entire process at one time, Rao is like Li Yanhong's purposeful visit, and after the oral agreement, he also carefully checked the content of the contract again.

Google is ready to draw up a share contract in a day or two, and the move is already fast.

The overall intention is basically reached, and the matter of joining Google is still a two-way street.

Meng Fanqi needs to rely on TensorFlow (TF), a deep learning framework that Google is about to develop, and the current way of implementing the code is too inconvenient.

Google's massive computing resources, and later the tensor computing unit cluster, are also indispensable tools for Meng Fanqi to achieve various innovations in the future.

Thousands of computing units, whether it's Nvidia's GPUs or Google's TPUs, are always more comfortable in the big factories.

If he builds his own cluster, not to mention the cost of hundreds of millions of dollars, it will be difficult for him to afford the wear and tear of supporting the continuous operation of these equipment.

So at this point, Meng Fanqi must join Google. Google has also accepted the way of sharing, but the specific division method needs to be carefully evaluated.

Now that the general direction of the entry has been decided, there will naturally be some bragging in the remaining time.

Don't underestimate this part, especially before the contract is officially signed. The pie is well spent, and the budget is cut to the foot.

It's really crippled for you, and it's very possible to take people down with a low budget on a low budget.

"It's been almost fifteen years since I joined Google, where I witnessed the rise of the internet, and now I think I'm leading the rise of artificial intelligence."

Now that we're down to business and discussing our vision for the past and the future, Jeff would like to share some of his and Google's plans and blueprints for the future of AI.

"Helping computers recognize objects, understand speech, and even have conversations that used to seem like a fantasy is now becoming a reality."

"In the last five to 10 years, computers have rapidly developed the ability to [see], and, in terms of your latest work, have skyrocketed to the human level."

Technology in the AI era is evolving too fast, which is the core reason why Jeff is willing to spend money to recruit talent.

"Google now has many scenarios where we want to develop AI technology, and we want to translate more than 100 languages to each other so that people can communicate better; We need to intelligently analyze medical images to predict and diagnose diseases more accurately. Among all these applications, in fact, the core is two things, algorithms and computing power. ”

Jeff's summary is very concise, modern AI is mainly based on the old algorithm of neural networks, and if you put aside the computing power to talk about AI algorithms, it is completely empty.

"Google is determined to build the world's most powerful computing platform, and we will definitely maximize the value of excellent and intelligent algorithms."

Meng Fanqi has no doubts about Jeff's determination, which is why he chose Google in the first place.

"The meaning of computing power is relatively pure and easy to understand. But there are too many meanings of algorithms, in fact, I personally think that the design of the network structure itself is not the focus and core thing.

If you really want to change the world, you need a framework and platform that is simple and easy to use, easy to deploy, and optimized and accelerated in terms of the data type of the operation. ”

At this stage, the industry pays great attention to the design of neural networks, which layer is designed and what operation is better.

During this period, the benefits of doing so are also huge, such as last year's AlexNet and this year's DreamNet, both of which have been terrifyingly improved.

However, in Meng Fanqi's view, the structure in the later part of the AI era has not changed much, and the most important thing is to vigorously produce miracles, plus he knows very well what task is better to use what structure, and the structural design is too simple for him.

"When the competition comes to the end, the training technology of large models and massive high-quality resources are more critical."

Jeff and Hinton secretly exchanged glances, feeling weird.

Originally, Jeff came to show this undergraduate, who was still in contact with research in school, Google's ambitions, what multi-field blossoming, the largest computing platform and so on.

How do you feel that this kid has such a clear understanding of the main pain points of the AI industry, and it doesn't look like he is doing research in an ivory tower.

The main purpose of AI research in academia is to verify a certain conjecture and improve specific indicators.

Industrial AI is more pragmatic, how to implement it requires fewer resources, how to make the model faster, and how to deploy it to different devices.

The two sides often look down on each other, and the academic community feels that the industry is just a wage earner doing dirty work, and there is no innovation and breakthrough. The industry thinks that the academic community will write papers and brag, and everyone will not use the things they make.

Jeff and Hinton can be said to be representatives of industry and academia, and even when Jeff was studying, his graduation thesis was in the direction of industry, parallel training of large neural networks.

At that time, it was only 1990, and Jeff had already begun to study the core technology for 2023, the training method of large models.

"I have to say, I thought you would be an academic person who has made breakthroughs in algorithms in a row." Jeff's expression was surprised and surprised, "I didn't expect your thinking to look at the problem and the needs of our industry are very consistent. ”

Jeff has been in contact with many outstanding scholars, and even Hinton has the inertia of academic thinking, so inside Google Brain, Hinton is not involved in any management and decision-making work, and is only responsible for academic research.

Maybe this time, I'm not just hiring a brilliant algorithm researcher, who may also be able to help me with the company's AI strategy.

Jeff had a vague hunch.

He was the technical backbone of Google for more than ten years, and he was not involved in many management projects. But the direction of AI is something that he strongly supports and promotes by Ng Enda, so many things in this area are led by him.

As a leader, Jeff likes a different perspective and likes new things.

For example, neural networks and AI, although he also studied it in the 90s, but after that, his work at Google was more about architecture, search, and advertising.

In fact, I haven't updated any AI knowledge.

It wasn't until 2011, when Ng collaborated with Google that he suggested to Jeff that things were changing rapidly and that Google should focus on AI technology.

Jeff embraced the change very quickly, and could even say that he was inherently interested in this kind of potential solution that he was not familiar with enough.

Once he understands the solution to the problem, he will lose interest.

Meng Fanqi's idea of AI strategy is different from his own, but Jeff is even more pleased, and secretly said in his heart, this time he really picked up a treasure.