"Do you have any other questions, Mr. Hu?" Meng Fanqi grinned, he had long noticed that Xi Xiaohu was deliberately examining him.

He was well aware that Xi Xiaohu was a leading figure in Google Shanghai's advertising, ranking at least second in command if not first.

If he didn't completely dispel Xi Xiaohu's doubts, it would be difficult to proceed with the work in the future, and subsequent progress and testing would be very challenging.

It's important to note that Google tests various methods thousands of times each year, and in the end, the algorithm still needs to rely on others.

Moreover, Xi Xiaohu's suspicion at the moment was understandable, not malicious.

When Xi Xiaohu heard himself being called "Mr. Hu," his expression stiffened, feeling that this kid was doing it on purpose. Why did this title sound so silly? It's like a tiger's growl.

Meng Fanqi, on the other hand, looked at him innocently. What could he do? He couldn't call him Mr. Xi, that would be taking away the title of future leader from old Biden.

"It seems that Mr. Meng is indeed very familiar with our overall strategy, I just don't know what improvements you plan to make." Xi Xiaohu had no objections to Meng Fanqi himself. If the other party could really make a breakthrough, it would be a great achievement for him.

Especially in this R&D process, it seemed to be carried out at Google Shanghai, and he was bound to become the person in charge and the gatekeeper.

Naturally, he would get a share of the credit.

By this point, Xi Xiaohu had basically put away his suspicions. At least the young man in front of him was clearly not here to cause trouble, and the lower limit of his skill had been guaranteed.

Now it remained to be seen what his upper limit was.

"My idea is to optimize mainly from the perspective of mobile devices and the explanation model of artificial intelligence." The first two algorithms that Meng Fanqi started with were methods iterated into Google in 2015.

We are "pawread.com", find us on google.

There was a certain distance from the current year, and he confirmed that this part of the internal work had not yet begun, but it was not too far off. The effect would be more predictable.

"The former is aimed at optimizing mobile web pages. It can be foreseen that in the future, more and more people will use mobile devices to search the internet. I believe that a method to evaluate the display effect of web pages on mobile devices can greatly enhance the search experience for mobile users."

At the end of 2013, browsing many web pages on a mobile phone was still very uncomfortable, with inconvenient zooming and many functions not supported.

However, in search engines, there were still very few that paid attention to this problem and incorporated it into web page rankings.

But just a few years later, PC-based web pages were already being ignored, and mobile traffic had increased tenfold.

It was a huge waste of search resources to push websites with poor reading experiences to mobile devices.

After listening to Meng Fanqi's detailed technical methods, Xi Xiaohu's pupils shook, completely suppressing any suspicion and leaving only amazement.

"Your approach is quite clever... by judging based on specific visual requirements, you can confirm that the proportion of web pages is indeed favorable to the current screen ratio of the user."

As an expert, Xi Xiaohu instantly realized the value of this method.

Although it did not directly generate revenue, this kind of care and attention to user experience was more appreciated by him as a technical leader than revenue.

"This direction can also include a variant, which is to prioritize websites that load quickly on mobile devices." The patch that Meng Fanqi made was a new judgment added by Google in 2018.

Mobile networks are not as stable as PC networks, and web page loading speed is also a major factor affecting the experience.

Xi Xiaohu was delighted after hearing the core content of this patch version. Previous suspicions vanished into thin air and were never mentioned again.

He put his arm around Meng Fanqi's shoulder and said, "The two things you've come up with are worth a year's worth of work for some teams!"

"Mr. Hu, aren't you going to ask any basic questions?" Meng Fanqi took the opportunity to tease.

"Hey! What kind of talk is that?" Xi Xiaohu waved his hand, "I'm just testing you casually!"

The two laughed, and the subtle tension between them was resolved.

"Another direction is the introduction of the new AI version." Meng Fanqi naturally did not forget his old profession. The judgment of mobile adaptation is not related to AI, but this explanation model can introduce more factors to comprehensively judge the relevant results of a search.

This is a very important technological breakthrough, and Meng Fanqi will also develop the Transformer and BERT (BERT) methods next year, which is the T of ChatGPT.

At that time, this explanation model will continue to be upgraded.

Even if Meng Fanqi leaves Google more than a year later, this algorithm will continue to iterate.

It can even be said that it is a very low-end version of Bing's Miss (GPT-4 core), with a certain language understanding ability sufficient to help search engines better understand user search queries.

However, it does not yet have any intelligent functions that resemble human conversation.

At this stage, Xi Xiaohu's understanding of AI is not so comprehensive. The language models at this time are mainly based on time-series models and are relatively old technologies.

Therefore, he still knows a little: "Time-series language models may indeed understand sentences well, but using them on such a large scale, will it be too slow?"

The so-called time-series model requires the calculation of a moment to be related to the context before and after, and cannot be calculated independently, which greatly limits parallel computing.

Originally, one person doing a hundred questions is indeed very slow, but if a hundred people do a hundred questions, that is a hundredfold increase in speed.

But if the second question must have the answer to the first question in order to be done, then there is no way, you can only wait.

The so-called one core working, seven cores watching, is a similar situation.

In this context, even if Google has enough computing resources, it is still difficult to handle the computation time problem.

It's important to note that the delay in searching is much more fatal than the delay in opening a web page."Tiger, have you forgotten what my old profession is? When it comes to the parallel issues of language models, I certainly have ideas before proposing this plan." When it comes to recommended ads, Meng Fanqi's details are far less solid and steady than Xi Xiaohu's.

But when it comes to breakthroughs in the AI field, even a big shot like Xi Xiaohu is no match for Meng Fanqi.

And the three-pronged approach to recommended ads that Meng Fanqi has prepared this time, in addition to the first two focusing on mobile optimization and search term understanding optimization. The last one, the Fred algorithm, is a method specifically designed to punish, and its effect is immediate.

"My final algorithm strategy is to specifically search for low-value content, deceptive ads, and various membership links in the search results. By using the most severe punishment, it forces websites to lean towards high-quality original content and a superior user experience."

With this, the basic framework of Google's main recommended ad algorithms from 19 years ago has been established.

All that's left is for Meng Fanqi to develop and complete it.