It is normal for negotiations like this to not be completed in one go. Even after Li Yanhong's visit with clear intentions, the contents of the contract were carefully checked again.

Google is preparing to draft a profit-sharing contract within one to two days, which is already quite fast.

The overall intention has basically been reached, and joining Google is a mutual decision.

Meng Fanqi needs to rely on Google's upcoming deep learning framework TensorFlow (TF) because the current implementation method is too inconvenient.

A quick look at bit.ly/3iBfjkV will leave you more fulfilled.

Google's massive computing resources and the later tensor computing unit cluster are essential tools for Meng Fanqi to achieve various innovations.

Whether it's thousands of computing units, whether it's NVIDIA's GPU or Google's TPU, it's always more comfortable to take advantage of them in big companies.

If he were to build his own cluster, not to mention the cost of billions of dollars, he would have a hard time bearing the ongoing expenses to keep these devices running.

So in this regard, Meng Fanqi is bound to join Google. Google has also accepted the profit-sharing method, but the specific distribution needs to be carefully evaluated by Google.

Now that the general direction of joining has been decided, the remaining time naturally involves some bragging and promising.

This part must not be underestimated, especially before the formal signing of the contract. If the promises are too good, the budget will be cut.

If you really deceive them, adding some value and sentimentality, it is very likely to win people over with a low budget.

"I have been with Google for almost fifteen years. I have witnessed the rise of the Internet in this company, and now I believe that I am leading the rise of artificial intelligence."

Now that the business matters have been discussed, Jeff wants to talk about the past and future visions. He hopes to introduce himself and Google's plans and blueprints for AI.

"Helping computers recognize objects, understand language and speech, and even have conversations. These things used to be considered impossible, but now they are gradually becoming a reality."

"Just taking computer vision as an example, in the past five to ten years, computers have rapidly developed the ability to 'see.' And based on your latest achievements, it has already reached the level of human beings."

The technological development in the AI era is extremely fast, which is the core reason why Jeff is willing to invest heavily in recruiting talents.

"Google now has many scenarios where we hope to develop AI technology. We want to achieve mutual translation of more than a hundred languages for better communication among people. We want to intelligently analyze medical images for more accurate prediction and diagnosis of diseases. Among all these applications, the most crucial things are algorithms and computing power."

Jeff's summary is very concise. Modern AI is mainly based on the ancient algorithm of neural networks. If we talk about AI algorithms without considering computing power, it would be completely empty talk.

"Google is determined to create the most powerful computing platform in the world. We will definitely maximize the value of excellent and intelligent algorithms."

Meng Fanqi does not doubt Jeff's determination. This is exactly why he chose Google in the early stages.

"The meaning of computing power is relatively pure and easy to understand. But the meaning of algorithms is too broad. Personally, I think the design of the network structure itself is not the focus or core."

To truly change the world, we need frameworks and platforms that are simple and easy to use, easy to deploy, and optimized for computational data types.

At this stage, the industry is very concerned about the design of neural networks, such as how to design each layer and which operations are better.

During this period, doing so brings great benefits. For example, last year's AlexNet and this year's DreamNet had terrifying improvements.

However, in Meng Fanqi's opinion, the structure in the later stage of the AI era will not change much. The most important thing is to achieve miracles with great power. He also knows which structure is better for which task, so the design is too simple for him.

"When the competition reaches the end, the training technology of large models and the availability of abundant high-quality resources become more crucial."

Jeff and Hinton exchanged a strange look. They felt something was off.

Originally, Jeff was here to show this undergraduate student who is still studying in school Google's ambitions, such as blooming in multiple fields and having the largest computing platform.

But why does this kid seem to understand the main pain points of AI industry so clearly? He doesn't seem like someone who only does research in an ivory tower.

In academia, AI research is mainly to verify a hypothesis and improve specific indicators.

Industrial AI is more practical, focusing on how to achieve the desired results with fewer resources, how to make the model faster, and how to deploy it on different devices.

Both sides often look down on each other. Academia thinks that the industry is just doing dirty and tiring work, with no innovation or breakthroughs. The industry thinks that academia only knows how to write papers and make empty promises, and the things they produce are not useful to anyone.Jeff and Hinton could be considered representatives of the industrial and academic sectors respectively. Even when Jeff was studying, his graduation thesis was on industrial topics, specifically parallel training of large neural networks.

Back then, it was only 1990, but Jeff had already begun researching the core technology of 2023, the training methods for large models.

"I have to say, I thought you, who continuously make breakthroughs in algorithms, would be more academically inclined," Jeff said, his expression a mix of surprise and delight. "I didn't expect your way of thinking to align so well with the needs of our industry."

Jeff had encountered many outstanding scholars, even Hinton had the inertia of academic thinking. Therefore, within Google Brain, Hinton did not participate in any management or decision-making work, focusing solely on academic research.

Perhaps this time, the person I'm hiring is not just an excellent algorithm researcher. He might also be able to provide significant assistance in the company's AI strategy.

Jeff had a vague premonition of this.

He had spent the previous decade as a technical backbone of Google, not involved in many management projects. But in the field of AI, which he strongly supported Andrew Ng in promoting, he led many of the initiatives.

As a leader, Jeff liked different perspectives and fresh ideas.

Like neural networks and AI, although he had studied them in the 90s, his work at Google later focused more on architecture, search, and advertising.

He hadn't updated any AI knowledge since then.

Until 2011, when Andrew Ng collaborated with Google on a project, he suggested to Jeff that the situation was changing rapidly and Google should pay attention to AI technology.

Jeff quickly embraced this change. One could even say that he was naturally interested in potential solutions that he was not familiar with.

Once he understood the solution to a problem, he would lose interest.

Meng Fanqi's ideas on AI strategy were different from his own, which made Jeff even more delighted. He thought to himself, this time, I've truly struck gold.