Meng Fanqi has imagined scenarios related to rebirth several times. The first step was to launch the residual network to this venue.
In his imagination, this moment was the real beginning of his career.
But what he didn't expect was that the sense of smell of the technology giants was much sharper than his imagination.
Before this report on December 7, Google and Baidu had already signed contracts with themselves.
When Jeff finally signed the contract with Meng Fanqi on behalf of Google, he also felt a burst of pain.
This is one of the most generous contracts in the history of Google's talent hunt.
The share ratio is basically set at the highest level of 30% for 1-2 years, and then 10% for three years.
All are the most extreme benefits in the draft contract given by Jeff.
"This may cost close to 100 million US dollars every year." This is the worst plan in the minds of Jeff and the relevant directors.
They had no idea that Meng Fanqi's profits from Google in just 2-3 years would be a terrifying figure dozens of times greater.
On this basis, he relied on investment and entrepreneurship to become the youngest and fastest self-made billionaire in history.
This made him stand out among a group of old men in their fifties and sixties.
This is the value and power of technology without relying on anyone's investment.
"I have to say, you are the first person who has made me so entangled." Before the meeting on the 7th, Jeff finally signed the contract with Meng Fanqi at the Sydney venue.
He shook hands with Meng Fanqi with a wry smile. Normally, he would rather give up than give in to this situation.
But Hinton convinced him that Meng Fanqi's talent and academic taste in AI research were unprecedented in his experience.
For the company's future in the direction of AI, he used his position at Google to guarantee the board of directors.
"Now that the contract is signed, I am ready to carry out relevant research." Meng Fanqi is satisfied with this final version. Although his technology is very strong, there is no better platform than Google at this time in the world.
If he really messed up, it would be a very passive thing for him.
"Although I will be a few months late in arriving in the United States, my willingness to cooperate with Google is very strong. I already have some ideas that I need your relevant data and materials to verify."
Meng Fanqi doesn't know much about Google's advanced recommendation advertising algorithm after 2020, but he has still read the classic papers in the 16-19 time period.
These techniques require Google’s internal customer behavior data in order to train the models well.
This is why the more AI develops, the more it looks like a monopoly industry. What forms the threshold is not technology, but high-quality massive data and the terrifying computing power required.
"Of course there is no problem. Now that the contract has been signed, we will give you an internal account with corresponding permissions. You can access the corresponding internal data and materials, but be careful about leaks." Jeff reminded.
Once these internal things are leaked, the compensation required according to the contract will be considerable.
"The more common problems have relevant entry documents and guidelines. If you need help, you can discuss it with my students Alex and Ilya at any time." Hinton is still quite optimistic about Meng Fanqi.
Alix and Ilya? Starting to work with the Chief Technology Officer of ChatGPT?
This feels really good.
It takes a long time to graduate with a doctorate in the United States, so Alex and Ilya are basically the same age as before Meng Fanqi was reborn, and their communication is relatively smooth.
Especially Ilya, he is currently focusing on research on seq-to-seq related problems, that is, sentence-to-sentence tasks, such as translation and dialogue.
Since the later ChatGPT is the most popular and shocking product in the history of AI, all major domestic manufacturers are trying to copy this technology.
Therefore, Meng Fanqi is most familiar with the path of Ilya in the field of natural language.
In the previous brief exchanges, Ilya felt that Meng Fanqi was always able to point him out where he was confused, and he was instantly convinced.
"A genius is still a genius, it's completely unreasonable." Ilya is actually a proud man. "He obviously mainly does research on vision, but he understands natural language issues more deeply than I do."
Ilya was quickly impressed and secretly determined to communicate and cooperate more with Meng Fanqi.
"Can I first understand the process and verification indicators of Google's current recommendation and internal iteration of advertising algorithms?"
Google had already prepared an internal account. After Meng Fanqi logged in and looked around, many things were quite familiar.
For example, Google's code specifications were the blueprint for the code specifications of many major manufacturers in China, so Meng Fanqi was basically familiar with them.
As for some internal tools, although many of them have not been used by Meng Fanqi, they are also famous and heard about.
So most features come into play quickly.
In the code management warehouse, most of the content that has nothing to do with you does not have open permission for you. In fact, this does not matter, he will not be able to understand it if it is open.
For example, the source code of Google Chrome has been open sourced for a long time, but most people don’t have the ability to understand it.
The amount of code is huge and complicated, and even if it is a summary written by others, Meng Fanqi can't understand it.
His main focus is on several algorithmic directions related to AI. The progress within Google is still very fast. He must hurry up and develop a few fast-monetizing technologies.
Then we will slowly make breakthroughs in AI.
"Advertising algorithm and recommendation algorithm?" Jeff and Hinton both had question marks on their faces.
If image algorithms and language algorithms have a lot in common, recommendation and advertising are very different areas.
In the past two years, images have made the fastest breakthroughs, followed by language.
Meng Fanqi himself has made three major image achievements in the past few months.
Jeff was still counting on him to achieve great results again at Google Brain, but why did he rush to recommend ads?
Is it just for money?
That's right, it's just for money, no other reason.
Algorithmic breakthroughs in this area can directly extract more than 80% of Google's revenue, which is much more attractive than AI technology in other directions.
"Aren't you going to continue working on image algorithms?" Hinton was a little worried. He was a relatively pure scholar. Although he was troubled by money in the early days, after making tens of millions, he no longer cared about money and settled down to do academic work.
But at present, it seems that Meng Fanqi is obviously not such a person. He went for the advertising business, which makes the most money. Such a utilitarian mentality makes Hinton a little worried.
"No, I have a little idea about the recommendation algorithm, but I haven't had data to support it before. My focus is still on images and language."
Meng Fanqi only remembered a few recommendation algorithms and had to make them quickly to monetize them. In a few years, his level in this area will not be able to keep up, and he will not be able to continue doing it even if he wants to.
Hearing these words, Hinton felt a little relieved.
