"Does Mr. Hu have anything else to ask?" Meng Fanqi grinned. He had long discovered that Xi Xiaohu was deliberately inspecting him.
But he knew that Xi Xiaohu was the leader of Google Shanghai Advertising, not to mention the first, but at least the second.
If his doubts are not completely dispelled, future work will be difficult to carry out, and subsequent advancement and testing will also be difficult to go smoothly.
You must know that Google tests various methods thousands of times every year. In the end, if the algorithm wants to go online, it still needs to rely on others.
Moreover, Xi Xiaohu is now just understandably suspicious, not maliciously targeted.
When Xi Xiaohu heard that he was called Mr. Tiger, his expression froze. He felt that this kid did it on purpose. Why does this title sound so silly? It's so scary.
Meng Fanqi looked at him with an innocent face. What could he do? He couldn't call him President Xi. Wouldn't that be taking away the rare title of old Biden in the future?
"It seems that Meng Fanqi really understands our overall strategy very well, but I don't know what improvements you plan to make." Xi Xiaohu has no opinion on Meng Fanqi himself. If the other party can really make a breakthrough, it will be a big win for him. It's also a big achievement.
Especially this research and development process seems to be carried out in Google Shanghai, and I am bound to be the person in charge and gatekeeper.
Of course, you can get a share of the results.
When the conversation came to this point, Xi Xiaohu had basically put away his doubts. At least the young man in front of him was obviously not here to cause trouble, and the lower limit of his level had been guaranteed.
Next we have to see what his high opinions are and where his upper limit is.
"In my opinion, optimization mainly starts from the two aspects of mobile devices and artificial intelligence interpretation models." The two algorithms that Meng Fanqi started with are both methods that Google has iterated on over the past 15 years.
There is a certain distance from the current year. He confirmed that this part of the internal development has not yet started, but the distance is not too far away and the effect will be more predictable.
"The former is aimed at optimizing mobile web pages. It is foreseeable that in the coming years, more and more people will use mobile devices to search online. I think a method that can evaluate the display effect of web pages on mobile devices can Greatly improve the search experience of mobile users."
At the end of 2013, many web pages were still very uncomfortable to browse on mobile phones. It was inconvenient to zoom in and out, and many functions were not even supported.
However, few search engines have noticed this problem and introduced it into web page rankings.
After just a few years, no one paid attention to the PC-side web pages, and the traffic on the mobile side increased tenfold.
Pushing websites with poor reading experience to mobile devices is a huge waste of search resources.
After listening to Meng Fanqi's detailed technical methods, Xi Xiaohu's pupils were shocked and he completely suppressed any doubts.
All that's left is surprise.
"Your idea is quite clever...by judging specific visual requirements, you can confirm that the proportion of the web page is indeed conducive to the user's current screen ratio."
As an expert, Xi Xiaohu instantly realized the value of this method.
Although it does not directly generate revenue intuitively, this kind of care and attention to user experience is more appreciated by him as the technical leader than revenue.
"A variant can be added in this direction, which is to give priority to websites with fast mobile page loading." This patch made by Meng Fanqi is a new judgment added by Google in 2018.
The mobile network is not as stable as the PC network, and web page loading speed is also a major factor affecting the experience.
Xi Xiaohu smiled happily after hearing the core context of this patch version. The previous suspicions disappeared and were never mentioned again.
He put his arm around Meng Fanqi's shoulders and said, "The two things you have made are comparable to what some groups can achieve in a year!"
"Mr. Hu doesn't ask basic questions anymore?" Meng Fanqi also took the opportunity to joke.
"Alas! What kind of words are you talking about?" Xi Xiaohu waved his hand, "I can just test you!"
The two laughed, and the subtle points in front were revealed.
"The other direction is the introduction of a new version of AI." Meng Fanqi will naturally not forget his old profession. The judgment of mobile terminal adaptation has little to do with AI, and this explanation model can introduce more factors to comprehensively Determine the relevant results of a certain search.
This is a very important technological breakthrough. Meng Fanqi will also develop the Transformer and 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-profile version of Miss Bing (GPT-4 core), which has a certain language understanding ability that is enough to help search engines better understand users' search statements.
But it doesn’t have any smart-looking features yet, so it can’t really talk to users.
Xi Xiaohu’s current understanding of AI is not that complete. The language models at this time are basically sequential models, and they are all old technologies that have been around for a certain period of time.
Therefore, he still knows something: "The temporal language model may be good at understanding sentences, but will it be too slow when used on such a large scale?"
The so-called sequential model means that the model calculation at a moment may need to be linked to the context, and there is no way to perform the calculation independently, so it greatly limits the parallelism of the calculation.
Originally, it was really slow for one person to do one hundred questions, but if one hundred people did one hundred questions, that would be a hundredfold increase in speed.
But if the second question requires the answer to the first question, then there is nothing you can do but wait.
The so-called one core is doing the work and the seven cores are watching, which is a similar situation.
In this context, even if Google has enough computing resources, it is still difficult to deal with this computing time problem.
You need to know that the delay when searching is much worse than the delay when opening a web page.
"Mr. Hu has forgotten what my main line of work is? The parallel problem of language models. Of course I came up with this plan only when I had an idea." When it comes to recommended advertising, Meng Fanqi's details are far less solid and robust than Xi Xiaohu's.
But when it comes to breakthroughs in the direction of AI, a big boss like Xi Xiaohu is far from Meng Fanqi’s opponent.
Meng Fanqi has prepared three recommended ads this time. In addition to the first two focusing on mobile optimization and search word understanding optimization, Meng Fanqi has prepared three recommendations. The last Fred's algorithm is a specially punished method, and its effect is the most immediate.
“My last algorithm strategy is to specifically look for low-value content and various deceptive ads and various affiliate links in the search results. Use the most severe punishment to force the website to favor high-quality original content and high-quality user experience. .”
At this point, the basic framework of several of Google's main recommended advertising algorithms 19 years ago has been established.
Just wait for Meng Fanqi to complete the development.
