Trying Metaso: Why Its AI Search Feels So Fast and Structured

Jiang Wu · 2024-02-27

*Originally published in Chinese on February 27, 2024.*

Metaso, a Chinese AI search startup, has drawn attention for how its search feels in use. I have been trying it, and these are my guesses about what makes it different.

Guesses about what sets it apart

Speed and structural logic

Metaso responds very quickly, and the structure of its answers is very clear. It may have written a knowledge-graph framework into the structure in advance.

How results are presented

The mind map in the results has probably been tuned too. It is not hard technically, but it is friendlier to users.

Caching and scale

I am not sure whether it caches data. Once search volume reaches a certain scale, locally cached data can already meet most needs. When the user base passes a certain order of magnitude, 100 questions can be merged into one to get a standard answer, and that standard answer becomes the ruler for improving results from then on.

Collecting knowledge and generating content

Knowledge and answers in the public domain are easy to collect and standardize. Another simple approach is to break through within the same type of question and then use the structure to drive the content. A structured index is easier to derive than a general one, since similar data exists to refer to. What remains is whether the corresponding content can fill the gaps.

The index and whether its data is valid

The overall experience really is good. It probably does not translate Chinese to English, index, and then give results back. It indexes Chinese text directly. But it very likely retrieves and appraises the sources behind its index database. Another possible approach is to judge how valid data is by how often its sources are cited, and then analyze, classify, and build an effective feedback mechanism.

Problems it may run into

Token use and cost

My blind guess is that token use is very high. If it builds a local index in advance, it may be better. If it relies purely on the model for interaction, I believe the cost is not small either.

Control over what is indexed

I am not sure whether there is a better answer for a feedback mechanism that controls the indexed content.

Business model and where the profit is

Search costs far more than an ordinary tool. Is there enough money to last until the business model turns positive? There are plans to charge users, but for now I cannot see what its higher-priced mechanism is. It is probably still in development.

The search entry point and what it asks of users

A mechanism that starts from a question takes a large share of search entry points, but it raises the bar for users. That is a key factor in search.

Overall

It is worth a try. I hope more search-oriented products appear this year and make the experience better for users.

Home · All articles · Topics · 简体中文