Magi.com, a search engine, is featured by the knowledge map and ratings displayed in the first search results.

So I found this question:

What do you think of Peak Labs’ new version of the search engine Magi for ships you haven’t played yet?

And my answer was – “Magi was the first search engine that made me expect what my input would give me.”

The nature of search engines

The essence of traditional search engine is intermediary. In the beginning, this intermediary can mix well in many directions.

But the golden age has passed, and the mobile Internet has dismembered all that. Shopping agents? Open a shopping APP, a dining agency? Open a takeout APP, a taxi service? Open a taxi APP. Everything.

Where traditional search makes money, apps slice it up, leaving search for its most essential purpose: as an intermediary for media content (text, images, videos, etc.).

Some people may ask, search engines also have mobile apps, so can still occupy a part of the mobile market?

So let’s look at this from another dimension. That is, the traditional search engine APP will be robbed of the time slice.

The theory goes like this. Everyone has only 24 hours in a day. Then we can conclude that each person must use the Internet for less than or equal to 24 hours per day. We can then conclude that the total time spent using the APP is also limited, and the time spent using each APP is also limited.

Therefore, if the time spent using one APP increases, the time spent using other apps is bound to decrease without much change in the overall time spent using them. For example, if a college student spends 4 hours a day on his mobile phone, if he is addicted to Zhihu, he must spend much less time on weibo than Shuizhihu.

Then the question arises: How much time is the college student expected to spend searching APP on mobile terminal?

Now, Magi meets his needs for content searches, such as homework, which can be provided with an outline of mindless writing at the click of a button.

May I ask, will the time of this college student using traditional mobile terminal search APP decrease or increase?

The core of the above description is that every time an APP takes up a user’s time slice, the usage time of other apps will inevitably decrease.

However, search engines dominated by other types of apps will be used for a very long time. This usage time data is fed back to DAU.

We consider again from another dimension, search engine is never the end of all traffic. Even search engines have such a thing as Onebox (a onebox is a calculator that is embedded in search results if you search for a calculator, rather than a website that offers a calculator to click through).

Onebox allows traffic to settle. But mobile apps are also killing oneBox. Would you rather use the ticket-buying APP or oneBox?

It is obvious that the APP for buying airline tickets has an advantage, because the search for airline tickets is likely to be for buying airline tickets. So traditional search engines can only serve those who don’t want to buy a ticket but want to find information about the flight.

Do you suddenly feel familiar? Yes, Magi is for this scenario. Magi captures the header demand for traffic that is likely to settle in search engines.

At this point, the application scenario of the traditional search engine may be completely replaced:

• In phase 1, mobile Internet emerged and a large number of intermediary services were replaced
• In stage 2, DAU decreases due to macro competition with other apps
• In stage 3, the application scenario and use value of itself gradually shrink, and eventually the core business (mediating media content) is replaced by a vertically similar product (perhaps Magi), and then there is no more

That’s right, not “search is the foundation of the Internet and will never be replaced,” but “Big search is probably not going to be the foundation of the Internet.”

What did Magi do right?

Magi focuses on the very nature of users’ search behavior in the face of the mobile Internet, and makes a slight improvement – “Help you think.” When you want to know something or information, traditional search engines only show it to you according to the Page Rank of results. You need to judge the reliability of information by yourself, summarize key points, and select the information you want most. And Magi did that for you.

To say the least, even if Magi’s search results are dismal (and they are), that doesn’t stop it from being a tool that gives me perspective and information.

There may be an argument that Magi is just oneBox doing a better job. Other search engines can do the same. But perhaps the most important point is that traditional search engines, in order to make more money, even if they do similar things, still tend to bias their search results toward commercialization. It’s their revenue stream and it’s not going to change fundamentally.

This makes it impossible for a traditional search engine to become an application optimized for media content retrieval services. That’s my big assumption, of course.

Magi has created a new kind of need that allows you to copy (retrieve) your homework while also satisfying your and others’ needs for answers (see other people’s conclusions). In the past, you could only do this through UGC. For example, you could search zhihu for “How to evaluate XXX”. Now, you have an option: “What Magi says about XXX”.

And it’s not a conflict, it has the nature of navigation, but Magi also provides a measure of confidence, which is very important and what sets it apart from other UGC content. If your imagination is big enough, this is the prototype of the MGC (Machine-generated Content).

You’ve probably seen poems written in RNN, and I believe it’s entirely possible that one day we’ll see novels written by machines. Why, then, is it impossible for a machine to say something about a thing or a problem?

How’s Magi doing?

In my experience, Magi works best with proper nouns, but not synonyms, which can lead to confusion.

All search results, with the exception of the iconic OneBox shown above, require considerable refinement and are too narrow to satisfy the long tail.

This is a good result:

This is the bad result:

In terms of good results, the search keyword “CQRS”, Magi gives full play to the advantages mentioned before. In addition to providing search results, Magi can also provide reference information and confidence for the retrieved content, which makes me understand how CQRS is born and what purpose it serves. I can further understand CQRS according to this map.

However, the search results themselves still need to be optimized. This CQRS search results, Bing and Google results are the best, directly giving Martin Fowler’s authoritative article, which is PageRank’s advantage. Magi followed, giving Github a link. I wonder if Github’s weight is too high. The results from other search engines were abysmal. Become a direct victim of secondhand knowledge.

The bad results are obvious. If you search for “Nginx error 499”, the Chinese search engine only tells you how to avoid 499, but does not tell you why 499 is generated at all. The content is just a blind guess and experience talk, without argument or analysis, falling victim to secondary knowledge. Both Google and Bing have good results. I also did a Magi search for “nginx status 499”, no difference. On Google, I found satisfactory results in Chinese, English, Japanese and Russian.

Some eggs

I don’t know if the name Magi was chosen to express its meaning of “magic, wise man “. But I immediately thought of Magi, the supercomputer in EVA (later confirmed to be derived from EVA and Matthew):

Also, the confidence rating reminds me of ヒトログ (hito-log) in Luban III PART5, which can give confidence rating to each piece of information in SNS:

These two works are highly recommended. If you are interested, you can check them out.

The last

Anyway, I bookmarked the site. Expect Magi.com to get better and better.