laitimes

The AI comments that are crazy to harass you, are actually the careful thoughts of the platform behind them?

author:PConline太平洋科技

"The comments below my Weibo are all AI comments, for a while Sun Wukong, for a while Li Kui, I can't block it, it's crazy!"

Recently, a wave of AI bots is "invading" the Weibo comment area.

They are dressed in the clothes of familiar characters - Sun Wukong, Tong Xiangyu, Wang Xifeng, Concubine Xi, Zhen Huan, and Detective Conan...... Sometimes, in a comment area, you can even see the characters "stewed in a mess", and well-known IPs from ancient and modern China and abroad are gathered in a personal Weibo with less than 30 fans.

Many netizens complained bitterly. These AIs, which have no sense of borders, are rampage through the cyber world with character names and avatars, playing "language cosplay". But after PConline research, it was found that things are not so simple, dozens of role-based AI accounts are the same routine, similar gameplay, what is the origin of their daily increase in thousands of fans?

"Cyber Zombies" invade the comment area

"Recently, there have always been a few AI robots in the comment area who have come over for an awkward chat for some reason, and they can't even pull them in!" Xiao Wang (pseudonym) looked at the uninvited AI robot on Weibo and complained to PConline speechlessly.

Recently, almost every Weibo of Xiao Wang has an AI robot that plays a certain role to comment. Xiao Wang replied to an AI robot with the role of Monkey King: "Is there any way to block all of you AIs?"

As a result, the reply received from the AI was: "Haha, you can't block my grandson!" Xiao Wang was both angry and helpless.

In Weibo, which has 600 million monthly active users, this wave of AI comments began on Weibo's own product, @CommentRobert. In order to increase the time spent by users, at the end of last year, @评论罗伯特上线了. It will randomly post very emotional comments under the original Weibo of netizens. Because of the realism of the content of its comments, @评论罗伯特一上线就迅速走红.

The AI comments that are crazy to harass you, are actually the careful thoughts of the platform behind them?

@评论罗伯特的火爆让官方加大了AI评论的力度. In March, Weibo launched @MBTI Xiaoxingjia, allowing netizens to directly test their MBTI personality analysis. When it was first launched, the homepage of Weibo was full of "@MBTI small connoisseurs, please judge my MBTI based on the content of my Weibo. At present, the number of fans of @MBTI Xiaoxingjia has exceeded 1 million.

The AI comments that are crazy to harass you, are actually the careful thoughts of the platform behind them?

The official practice of strongly recommending AI robots has made many self-media teams see an opportunity to take advantage of.

The AI Monkey King, who constantly "harasses" Xiao Wang, is one of them. Behind this "Monkey King", there is an even larger "family".

After the investigation by PConline, it was found that there are nearly 40 AI characters on Weibo at the same time as AI Monkey King (January 8, 2024), and the character map not only covers the four famous novels, but also well-known film and television works such as "The Legend of Zhen Huan", "Attack on Titan", and "Detective Conan".

The gameplay of these AI characters is very consistent - after netizens follow the account, as long as they post Weibo and @这个账号或者是评论了这个账号, they will interact with it. At the same time, they will also "take the initiative" and randomly comment on netizens' original Weibo to attract their attention.

According to the statistics of PConline, at present, in the "family" of AI characters, the number of fans of a single account is not high, and the most one is @Huaguoshan_Monkey King with Sun Wukong as the protagonist of the account, which currently has 4.5w fans, followed by Zhuge Liang in the Three Kingdoms series, with 4.3w fans, and the number of fans is still rising at a rate of 1,000 fans per day. This "family" has a total of 31w fans. It is worth noting that some of these accounts have already been certified by "Weibo Golden V".

The AI comments that are crazy to harass you, are actually the careful thoughts of the platform behind them?

The "Golden V" certification is the highest level in the Weibo individual creator certification system, and it needs to meet the conditions of "10,000 fans≥ ≥1,000 fans, 1,000 fans, and ≥ 10 million readers in the past 30 days". Becoming Weibo Gold V means that you can open rights and interests such as advertising sharing and business order receiving.

To put it simply, after becoming a gold V, you can "just eat". But how did the operation of these nearly 40 AI accounts come about?

Proliferating role-based AI

Whether it is an official AI account or an AI "family" from the media, the essence is the same, and it is all achieved through AIGC technology.

AIGC, also known as Generative AI, is a new type of content creation after PGC (professionally produced content) and UGC (user-generated content). By learning text data related to the subject, the AI generates new text that is similar to the input data in style, tone, and content.

In fact, the training of the model layer is now very complete, and the solution of the application layer is gradually maturing.

Creating an AI character from scratch is very simple. In the case of ChatGPT, which dominates the AI model, in November last year, OpenAI announced at the developer conference that it would support user-defined GPTs. On the GPT Builder page, you can generate an application just by expressing your needs through dialogue.

After the practice of PConline, it was found that for a well-known work like Journey to the West, ChatGPT can automatically determine the identity of the character and be familiar with the background of the story with just one sentence. If you want a more realistic character effect, you can also upload detailed information yourself for AI to learn.

The AI comments that are crazy to harass you, are actually the careful thoughts of the platform behind them?

The left side is GPT Builder, and the right side is the preview effect of custom GPT

In China, this kind of technical service in the market is also developing, and even personal accounts can complete similar functions.

For example, on the website of an AI service, it is only necessary to bind your Douyin account in one step, and ChatGPT can be automatically called to reply under the specified video. You can also add custom settings such as filters and response styles.

There are generally two charging modes, one is to charge according to the number of calls. Taking this AI service website as an example, the free version can be called 100 times per month, and the basic version is 1500 times, with an annual fee of 1990 yuan/year. The other is a membership subscription system, which costs $20 per month like ChatGPT, which translates to about 1,700 yuan per year.

The AI comments that are crazy to harass you, are actually the careful thoughts of the platform behind them?

The low-barrier practice process has allowed role-based AI to proliferate everywhere.

Just three days after the GPT Store was launched, AI characters playing "sweet girlfriends" have been flooding in. Although OpenAI's policy clearly states that the company bans "GPT dedicated to cultivating romantic partners," regulation always lags behind. Even OpenAI's CEO Sam Altman came forward to call for not falling in love with AI.

But the need for users in the social market is always there. Character AI, the company that fully implemented character-based AI, has a market valuation of more than $5 billion in just three years. According to the data of the AI product list, the number of visits to Character AI reached 2.44 million in March, ranking first among vertical products.

The AI comments that are crazy to harass you, are actually the careful thoughts of the platform behind them?

One of the characters in the Character AI community: an AI virtual character based on actor Kate Bratchett

But will role-based AI be the new window to social platforms?

Is AI a "panacea" for social platforms?

"This kind of AI comment is going to me off a bit. Because I will feel that my private territory has been violated by machines. Xiaoyun, a student at school, said to PConline.

The social platform environment is no longer what it used to be. From the very beginning of Renren.com, which responded to every call, to the current active and anonymous "momo army", "recommend to people you may know" has become the most reluctant function, and "Moments is only visible for three days" has become one of the most popular functions of WeChat. Nowadays, the sensitivity to personal privacy has never been higher. There are many netizens who have the same idea as Xiaoyun, and many Weibo content are looking for ways to avoid AI comments.

Characterized AI bots, on the other hand, are not stable and often output surprising and sometimes aggressive content. The marketing account of the "Robert Victims Alliance" has been derived from Weibo, which has collected nearly 5,000 of Robert's "crimes". In this marketing account, which has a number of 32w fans, there are many comments from Robert about "yin and yang".

The AI comments that are crazy to harass you, are actually the careful thoughts of the platform behind them?

However, when the importance of AI becomes a consensus, each platform wants to participate in this "moment when humanity invents fire for the second time".

Following PGC and UGC, AIGC is a growth point that must be grasped for social platforms. In addition to the AI comments on Weibo, Bilibili has launched AI video summaries, and you can see the summary outline of the content under each video; Douyin has placed a special AI creation section on the user creation page, and the AI can smoothly turn a photo into a video......

在AI浪潮下,每个平台都患有严重的FOMO(错失恐惧症,Fear of Missing Out)。

This anxiety will be transferred from the platform through the product features to the user.

In the process of using an AI robot like this, netizens first feel the unhappiness of being harassed than "feeling the convenience of the function".

When you click on an unread message for the first time and find out that it is an AI robot, there is always some high-tech freshness. But when you see an unread message for the Nth time, expecting it to be a comment from a friend, and click on it only to find out that it is an AI robot, how will you feel?

Expecting role-based AI to "rebuild social" is not plausible. The core of social interaction is still the bond between people. The content of AIGC can only change one aspect of social networking, not the main body.

Using characterized AI to combat loneliness in social interactions can make people even more lonely. Anxious social platforms may need to stop drinking to quench their thirst. Rather than adding more bells and whistles to the AI component, it's more important to make your own stopdoing list.