Who is using DeepSeek to tell fortunes? I advise you not to...

Have you ever used DeepSeek to “tell fortunes”?

DeepSeek has been very popular since the beginning of the year. People have developed "100 ways to use AI". However, the most popular one is probably "DeepSeek fortune-telling" . Search "DeepSeek" on the Internet, and "DeepSeek metaphysical instructions" is listed in the drop-down recommended options. Clicking in, the number of likes and comments on these posts is simply staggering.

Image source: Screenshot from a social platform

What’s even more interesting is that DeepSeek seems to have a special preference for crystals in the process of “fortune-telling”, and frequently recommends various crystal ornaments to users as so-called “fortune-changing tools”. This also triggered an unexpected “crystal fever” at the beginning of this year.

So, the question is: What exactly is “AI fortune-telling”?

The essence of the large model is

Probabilistic Model

In fact, before DeepSeek ignited the "AI fortune-telling" craze, earlier AI products such as ChatGPT were also used as tools for cyber fortune-telling.

Image source: Screenshot from a social platform

But no matter which one it is, there is a fundamental logical paradox in the "AI ** fortune-telling" thing. The ** big model is essentially a probabilistic model that makes predictions through statistical probability, but it only imitates the language expression and analysis framework of the so-called "fortune teller" in terms of presentation.

Take the fictitious Gregorian birthday as an example:

Swipe up and down to view

DeepSeek can indeed calculate the so-called "Eight Characters" based on clear birth time information, which looks quite reasonable at first glance.

The so-called "fortune analysis" made by DeepSeek has the same content structure and expression as the common "fortune game" content.

What’s particularly interesting is that AI will also recommend “wearing red crystal” in the end. This is also the standard operation of many “fortune tellers” in reality - after analyzing the “fortune”, they always recommend some so-called “transfer items” to achieve commercial sales.

This "problem - solution" model makes the entire "fortune-telling" process appear more complete and practical, and is more in line with people's expectations of "fortune-telling".

The reason why many people think that DeepSeek's fortune-telling is very professional is actually just because its training corpus contains more and more sufficient relevant information, which can better fit some content in Chinese traditional culture, making the whole process appear more "convincing" and "immersive".

The psychology behind fortune telling——

Barnum Effect

If you use DeepSeek to "calculate" the same thing multiple times, you will find that... it gives a different answer each time.

AI products such as DeepSeek have a serious hallucination problem (you can review the previous article "How serious is AI in making up nonsense? Beyond your imagination!"). Especially when it is asked to speculate on the "fortune" of certain moments in the past, it often produces obvious nonsense.

The biggest reason why DeepSeek's "fortune-telling" seems to be "somewhat accurate" is that most of the predictions it gives are vague and cannot be verified . For example, the description in the previous picture: "This year is a year of 'requiring extra effort to achieve results'. You may encounter fierce competition or project pressure, but those who perform well will have the opportunity to be promoted or get important tasks."

This is when the famous "Barnum Effect" in psychology comes into play: give a description that is vague enough and universal enough for people to identify themselves with it.

The Barnum effect, also known as the Forer effect and the Astrological effect, describes a cognitive error in psychology, that is, people often believe that a general and general personality description accurately reveals their own characteristics. It is a typical cognitive bias behavior.

Trained AI big model products have “mastered” this “art of fuzziness” and are able to fabricate content that appears personalized but is actually applicable to the vast majority of people.

Image generated by AI

What is more alarming is that compared with human "fortune tellers", AI may be more "outrageous" in some aspects. Since many generative AIs are designed to meet user needs as much as possible, they tend to cater to the expectations of the questioner and give more irresponsible or even completely fictitious "predictions".

Therefore, DeepSeek is at best a performer who knows how to "get into the role" better, and an imitator who is more proficient in the traditional Chinese cultural discourse system.

There’s nothing wrong with enjoying “AI fortune-telling” for fun, but be careful not to take predictions that have no scientific basis as truth. You may make irrational decisions , such as blindly buying so-called “lucky crystals” or adjusting important life plans based on AI’s suggestions.

What is more worthy of vigilance is not to let "AI fortune-telling" encourage the spread of superstitious ideas. In today's rapidly developing science and technology, we should look at the world more rationally, but "AI fortune-telling" may make metaphysics popular again in a new look of "technology blessing".

No matter when and where, we should remember that AI-generated content is always just a prediction with the highest probability, not the truth, and AI is even less likely to predict fate. The real wisdom of life lies not in seeking external predictions, but in cultivating internal judgment and the ability to deal with uncertainty.

Planning and production

Author: Tian Wei AI tool researcher

Review丨Yu Yang, Head of Tencent Xuanwu Lab

Fan Chunlei, Associate Researcher, Institute of Psychology, Chinese Academy of Sciences

Planning丨Lin Lin and Yang Yaping

Editor: Yang Yaping

Proofread by Xu Lailinlin