I am the only one who practices magic: I practice magic in the city

Chapter 175 Crazy Pomelo Technology (Second Update)

3061年12月30日，加州时间21:50，湾区圣克拉拉山景城的一座看起来并不起眼的红色三层建筑中依旧灯火通明。

A thin-looking man in his thirties wearing a blue plaid shirt is concentrating on looking at the computer screen on the left. The codes he keeps typing out are displayed on the computer screen on the right.

Almost all the buildings visible to the naked eye here are owned by Google.

This seemingly inconspicuous building is the new department that Google just established more than four years ago - Google Brain.

A few months ago, Google just acquired DeepMind, the company founded by the super genius Demis Hassabis, who was admitted to Oxford at the age of 15 and independently developed the hit game "ParkSim" at the age of 17. This has brought some pressure to Google's brain team.

Especially in the past month or two, David from DeepMind seems to have made a very big breakthrough in the Monte Carlo tree algorithm.

Gugou gave them a lot of resources, which made Gugou, as Gugou's own son, feel even more pressure.

But it's okay, because DeepMind and Google Brain are conducting research at two different levels. DeepMind is more application-oriented, while Google Brain is more basic research-oriented, so the cooperation between the two parties is greater than competition.

"Jeff, Jeff! Professor Fang, Yu and Yortz from Youzi Technology published another paper on arXiv, "Long-range Dependencies in Sequences Based on Self-Attention Mechanism". It's amazing, this is the third blockbuster paper this month!"

A thin and small Asian man ran to the workstation of Jeff Dean, who was wearing a blue plaid shirt, holding a pad and handed it to him in English with a slight accent and a little excitement.

"Oh my god, I can't find out who this Professor Yoelz is, and as for Fang Yu, the information I can find shows that he is actually a college student studying finance at Da Zhou National University, not computer science or mathematics. This is really crazy!"

“You must read this paper ‘Based on Self-Attention’. It is almost the same as what Lucas’ team is working on. But it is obvious that Youzi Technology is ahead of us.”

"Please stop for a moment and read this article. This paper can save us at least six months of time. What are you...Oh my God, what is this?"

The thin Asian man glanced at Jeff Dean's computer screen and was stunned.

Jeff Dean had a wry smile on his face: "Li, maybe they are not only ahead of Lucas' team's 'addition' research."

"Is this a training frame?" The thin Asian man put his hand on Jeff Dean's mouse unceremoniously and started scrolling up and down.

“Yes, they call it yuzu 0.1. I just tested it and it’s at least 0.5 times more efficient than our TensorFlow 30.”

"Version 0.1 of the framework actually has libraries? And 12 tools have been released?" The Asian man surnamed Li was dumbfounded.

"Have you asked Wu? What did he say?" Jeff Dean pinched the bridge of his nose. He has always been very calm, rarely frustrated and rarely loses his temper, but today, he felt deeply frustrated for the first time.

The man surnamed Li looked away from the computer screen and shook his head. "I have asked, but Wu has never heard of this company. He is now looking for someone to ask about the details of this company. The registered address of this company is at the National University of Da Zhou. According to him, computer science at the National University is not a very strong subject in Da Zhou."

The Wu they are referring to is Andrew Ng, the former director of the Stanford Artificial Intelligence Laboratory who co-founded Google Brain with Jeff Dean more than four years ago.

The thin Asian man named Li is Li Yueguo, another founder of Google Brain, and Ender Ng is his doctoral supervisor.

Including Li Yueguo, nearly half of the core technical team of Google Brain are students of Ende Wu.

It can be said that without Andrew Ng, there would be no Google Brain today.

And Andrew Ng also came to Da Zhou last year and joined Qianxun, becoming Qianxun's chief scientist and taking full responsibility for Qianxun's AI research. In just over a year, Qianxun's development and research in many key areas of AI have surpassed Google.

In fact, before 3061, the development of artificial intelligence in Da Zhou even surpassed that of Mai in many aspects.

In terms of the number of journal papers mentioning "deep learning" and "deep neural networks", Da Zhou had surpassed Maiguo in 3059, and the quality of the papers was not bad either.

This is also why Wu Ende chose to go to Qianxun. On the one hand, Qianxun was very sincere and Robert invited him personally; on the other hand, it was also because Da Zhou itself had shown great talent potential in this regard.

"This training structure is much better than Qianxun's paddle. Wu should be very interested in it." Jeff Dean smiled.

"It seems that in addition to Qianxun, we may have another rival." Li Yueguo nodded in agreement, and suddenly thought of something, turning to look at Jeff Dean, "Do you want to acquire this company now? You can ask Selina." Jeff Dean spread his hands helplessly, "We can't decide which company to acquire, but I tried to send an inquiry email to this company a few hours ago, asking if they are willing to accept Google's acquisition."

Li Yueguo became excited: "Any reply?"

Jeff Dean clicked on gmail: "See for yourself."

Li Yueguo leaned over. Jeff Dean had written several thousand words of technical discussions and acquisition invitations below, but the other party's response was just one word.

No.

There is not even a signature.

"Forget it, Wu will have a bigger headache than us. By the way, we can use this Yuzu architecture to test the NMT training..."
-
Andrew Ng did have some minor headaches, but they were not caused by YUZI Technology, but by internal issues.

After he joined Qianxun last year, he did have a honeymoon period, had very pleasant interactions with his superiors, and did a lot of groundbreaking work.

But as time went on, he realized that he still wanted to keep some things simple.

Who would have thought that the boss of this company really doesn’t always have the final say?

Fortunately, the team is very motivated. After optimizing paddle, they have now developed deepbench. It is estimated that in one or two years at most, Qianxun's entire AI business will be mature.

Deepbench is a platform for evaluating the deep learning performance of hardware platforms. In layman's terms, it is Master Lu used to evaluate the deep learning performance of hardware.

Moreover, he just married a beautiful wife last year, and his relationship is at its best now.

Therefore, Wu Ende was in a pretty good mood.

Until he saw the Yuzu Framework 0.1 released by Yuzu Technology on GitHub.

The framework uses the Apache 2.0 license, which means it is open source but not completely free software.

What surprised him secretly was that the performance of this framework was extremely powerful. It was not only more powerful than Qianxun's Paddle, but also much more powerful than Google's TensorFlow0.5.

If the powerful performance could be the result of a lucky coincidence for this small company, then the fact that Youzi Technology is constantly improving its platform ecosystem has made it completely uneasy.

In just one week, Youzi Technology has uploaded 12 sample codes, models and libraries of various sizes. Each one is very practical and closely integrated with the Youzi framework.

Moreover, in addition to Youzi Technology, there are now other developers uploading sample codes and enriching libraries.

This framework was just released one day ago, right?

Wu Ende frowned and raised his hand to send a message to his boss: Robert, I hope to acquire a company, Youzi Technology, and here is the company's information.

Although Qianxun has many problems, Wu Ende still trusts Robert.

The day after he joined Qianxun, he asked Qianxun to buy a thousand GPUs. Within 24 hours, these GPUs appeared in his laboratory.

But if he uses Google, after submitting the application for approval, it may take weeks or even months before he can apply for the corresponding computing resources in the cloud.

"Ding." The message on the phone flashed.

"Okay, I'll discuss this with Melissa."

(End of this chapter)