(Ps: The new book that was cut off has been updated with chapter content. It is impossible to put this one. It is not that crazy. It’s just that the name of the search engine refers to “Spirit Realm Search”, because I thought of several names but didn’t think it was suitable. Nowadays, artificial intelligence and GPT
So hot, 2013 happened to be the year when biometric technology and neural networks began to explode. Originally, the plot of AI was moved from the outline of this book and was not expanded on in this book. The new book has been released and it has not been expanded there yet, so I adjusted it back.
But the content of chapters that have been updated will never be posted a second time.)
——
When Chen Yu said this, he turned off the PPT on the big screen of the conference, turned to look at the crowd and said: "In the construction of a large AI language model, a major focus of the subsequent work is data feeding. Although the underlying algorithm architecture is not based on natural language processing,
NLP, but there is no doubt that it is also affected by it. The next data feeding is to first crawl all the existing public data on the Internet and the existing data generated before 2012, both domestic and foreign."
This is tantamount to using most of the knowledge accumulated by all mankind over thousands of years for free.
An engineer attending the meeting asked: "Approximately how much data should be crawled every day?"
Chen Yu said concisely and to the point: "About 5 PB."
5PB?
Everyone was shocked. The amount of data in 5 PB is converted into more than 5 million GB. The amount of data processed by the Internet giant Google every day in 2008 was about 20 PB, which is equivalent to a quarter of its data.
Processing volume.
This is a big project, and it consumes a lot of computing resources. In other words, it burns money. The Internet fee alone will cost a lot of money.
Chen Yu turned around and said: "In addition, Sogou Search has developed a new version for a major update. After the new version is launched, it will be renamed Lingjing Search. When users are completely accustomed to it, the name Sogou Search will be abandoned."
In the early stage of the launch of the new version, the name of Sogou Search will still be retained. If the user enters Sogou Search Engine, it will jump to the display of Spirit Realm Search, and there must be a mark that this is the renamed Sogou Search.
Tell old users that the old dog is still the same old dog, the name has just changed.
Now it is definitely not possible to directly search this name using Spirit Realm. That will lose many old users. The name Sogou still needs to be maintained for a while. When the popularity and influence of Spirit Realm Search covers Sogou, the name can be completely abandoned.
At this moment, Fang Hong, who was listening in, crossed his legs and listened without saying a word.
Chen Yu continued: "For the new version of Lingjing Search, in short, it is simpler, more accurate, and more comfortable. The day Lingjing GPT matures, it will be connected to Lingjing Search, which is an important factor in subverting contemporary search engines.
.”
"Perhaps it may not be Lingjing Search that subverts Baidu or even Google, but there is no doubt that the advancement of AI technology will inevitably change the way people access information."
“The way I imagine it is that the search bar of the browser is replaced by human AI, and as I type, the AI automatically completes my idea or question and provides me with the best answer, which may be a website or product.
For links, the AI uses the old search engine backend to gather relevant information and connections and then summarizes them for me.”
"This method of subversion is like letting a professional researcher do the work, but the AI will complete it immediately, while humans will take several minutes or even longer to complete it."
When users search for content by themselves, they have to filter and search, which is sometimes very time-consuming. However, it is different with the help of AI. Just ask the AI directly, and it will provide the best answer to the user in seconds.
.
To achieve this effect, it requires super huge computing resources and AI that is "smart" enough or able to understand human "language" more accurately to provide accurate answers.
Obviously, first of all, AI needs to master all the knowledge accumulated by humans over thousands of years.
Let’s crawl that data first. This is one of the prerequisites for being able to quickly give accurate answers to any questions raised by any user.
At this moment, Chen Yu looked at everyone and said: "As we all know, search advertising uses keywords to attract traffic, which is very accurate and has high conversion effects. However, search advertising itself relies more on users' spontaneous search behaviors, resulting in certain limitations in its coverage.
Although the monthly search volume increases and decreases, it is generally limited.”
"How many visits to the search page account for the total visits?" Chen Yu said and looked at one of the people in charge attending the meeting. He was an employee of the former Sogou. Now he had arrived at the company headquarters and heard Chen Yu ask him
He immediately answered: "The proportion is about 8~11%."
Hearing this, Chen Yu nodded and continued: "This means that when you are operating an unpopular product, or when you want to get more advertising coverage, search advertising may not be able to give you much help.
So we need personalized display advertising.”
"If search advertising determines what ads will appear based on the user's search behavior, then personalized display advertising 'guesses' what the user is interested in and recommends what ads."
"Personalized recommendations require the support of new technologies. In addition, user portraits are also very important, which requires the accumulation of rich user data, such as what users often search for, preferences and other factors."
“We’ll talk about the technical support for personalized recommendations later. Let’s talk about the placement of personalized ad display first.”
"The new version of Lingjing Search is divided into two parts: left and right entries. The left entry shows the search keyword content, and the right part displays the 'guess' personalized display ads that the user likes. Each page contains a maximum of 5 advertising spaces. If there are less than 5,
The ads will be displayed on every page, and ads more than 5 will appear in sequence."
The experience of today's search engines is really hard to describe, and it takes a lot of time for users to search for the content they are looking for.
Because they forcefully "guess" you like to throw display ads at you, and domestic search engines don't mark it as an advertisement, so when you click on it, you find it is an advertisement.
As for the bidding ranking that has been criticized and complained by users, it goes without saying.
In the new version of Lingjing Search, according to Chen Yu's requirements, search ads and display ads must be separated. The left side is the terms produced by the user's own search, and the upper part on the right is "guessing" the user's favorite display ad content.
As for the lower right half, it is left blank. There is no content for the time being. A hot search list will be added here in the future.
There is no doubt that reducing the time cost for users to search for content is an improvement in user experience, and the same is true for being able to display the content that users are looking for more accurately. Of course, this is the core technical issue.
In addition, in order to improve user experience, Chen Yu also requested to increase restrictions on the advertising content of Lingjing Search advertisers, abandon many low-quality ads, and support high-quality advertising content.
This will definitely make a lot less money, and many financial backers will simply disappear.
The profitability of the revised Lingjing Search may decline significantly compared to the original Sogou Search.
At this moment, Fang Hong, who was silently attending the meeting, couldn't help but nod to himself. His guess was correct. Chen Yu was indeed not building a search engine on a whim.
Chen Yu has a long-term vision and does not care about this small profit. This coincides with Fang Hong, and he does not care about this small profit.
A product that is more powerful and has better experience will definitely not be bad at making money, but it will need to endure the situation of not making money or even losing money in the early stage. There is no problem with quantitative capital. Chen Yu can make money in the capital market himself.
If that's not enough, the parent company Qunxing Capital and the silent big BOSS sitting next to him will take action.
