Von gruseligem Video bis hin zu gefälschten wissenschaftlichen Artikeln verwandelt KI das Internet in einen riesigen „Müllplatz“.

This article is from the WeChat official account “Almost Human” (ID: almosthuman2014). The author is someone who follows AI. It is published by 36Kr with authorization.

Who could have thought that a video generated by AI would receive 252 million views on Instagram, with 3.257 million likes alone.

In the video, a fat woman in a bikini clumsily steps onto the railing of the boat and suddenly jumps into the sea. Instantly, a huge splash is created, and at the same time, the boat loses its balance due to her weight and movement and capsizes into the sea.

Keen - eyed friends have also noticed many signs of AI at play. For example, the man watching the show by the boat can turn his head 360 degrees like an owl, and the boat tilts in an unreasonable way.

The video touches the audience's funny bone with exaggerated plots, but it also has the suspicion of body shaming. Due to the huge traffic it has earned, the blogger has made many similar videos.

If you dig deeper into the blogger's other works, most people will feel a strong sense of discomfort, like getting goosebumps all over. There are divers with human bodies and pig faces, interviewees in zombie images, and strange and grotesque unknown creatures... One is scarier than the other. (The AI horror content is too high. To avoid causing discomfort, the video is not shown 😂)

Now, when you open almost any social media platform, several strange and eerie AI videos will pop up. Recently, the popular one is the AI - generated "cannibalism" video.

In it, all kinds of food not only grow human faces but also "eat" each other. Lemons, oranges, steamed buns, broccoli, blueberry muffins, etc. all close their eyes and smile, waiting to be fed by a spoon.

From the Instagram blogger reallyweirdai

The production process is too simple. With just one prompt, Veo3 can generate it:

A surreal orange with a human - like face rests on a sunlit kitchen counter, its textured skin illuminated by warm morning light; a spoon carrying a juicy orange wedge slowly approaches its closed mouth, and as it nears, the orange's lips animate and gently part to accept the slice. Chewing with an oddly natural motion before returning to stillness. The background remains softly blurred, drawing focus to the bizarre and whimsical act; the camera is static in a tight close - up shot with shallow depth of field, capturing fine textures and subtle movements; lighting is natural and directional, casting soft highlights and shadows across the scene, creating an uncanny yet playful atmosphere with dreamlike surrealism.

If you think this kind of video looks a bit cute, then the following one is definitely disgusting.

The video named "You Are What You Eat" once went viral on Reddit. It shows people made up of food eating these foods. For example, a sushi - made person is gobbling up sushi, or a yogurt - made person is gulping down a strawberry yogurt cup... If you add sound, it's really creepy.

The above video is from the blogger Bennett Waisbren. His inspiration comes from food - eating live - streaming videos, highlighting modern people's desire for gluttony in a dark style. Although the intended connotation is profound, it gives people a strong sense of discomfort visually.

Netizens' evaluations of this kind of video are also polarized. Some people think it is "creative" and that "you can earn huge advertising fees with almost no effort", while others think it is "too disgusting" and "a nightmare".

This disgusting feeling is actually because it touches the "uncanny valley effect", that is, when an object looks almost human but not completely, it will arouse people's disgust and fear.

The food in the video is given very realistic human facial expressions, including closing eyes and smiling, chewing, etc. When they show human - specific behaviors, it will make people feel uncomfortable.

These strange and eerie AI videos are very popular on social media platforms. Behind them is a blatant traffic logic.

The platform's algorithm favors content that can quickly catch people's attention and trigger interaction. Strange and exaggerated videos can naturally arouse the audience's curiosity, leading to more clicks, comments, and shares. This trend of pursuing traffic and short - term benefits encourages creators to constantly cater to the audience's curiosity and make more extreme and non - traditional videos.

Coupled with the rapid development of AI tools and low - cost creation, creators can generate highly realistic videos with just a short instruction. The creation process has become easier and more efficient, which has exacerbated this phenomenon to a certain extent.

More seriously, the "junk" generated by AI is not only reflected in entertainment videos but has even penetrated into the academic field.

Researchers at the University of Borås in Sweden once said in a study that in Google Scholar, a literature indexing database, they found hundreds of articles suspected to be "fabricated" by AI. Using AI's low - cost "cheating" method to easily create and spread false scientific information not only erodes the credibility of academic platforms but also consumes the real value of science with this low - quality pseudo - science.

Another study investigated scientists' peer reviews. At one conference, the frequency of using the word "meticulous" in peer reviews was 34 times that of last year, the frequency of using "commendable" was about 10 times that of last year, and the frequency of using "intricate" was 11 times that of last year. These kinds of words are among the favorite buzzwords of large models like ChatGPT.

In other words, researchers in the AI field are increasingly relying on AI tools to assist or even do peer reviews on their behalf, especially when the deadline is approaching.

Some AI - generated content is obviously fake. For example, last year, a medical journal paper shocked scientists. The author used Midjourney to generate a picture of a cartoon mouse with genitals. Not only was the anatomical diagram completely wrong, but it also contained a lot of scribbled text. The fact that such an obviously wrong paper could pass peer review and be published in an academic journal shows the seriousness of the problem behind it.

Scientists' reputations rely to a large extent on their publication records, such as the number and frequency of published papers and whether they can be published in top - tier journals. But this mechanism also gives some people an opportunity. They use AI to generate low - quality articles, seriously harming the purity of knowledge and the fairness of scientific research.

Whether it is the AI horror videos on the Internet or the low - quality AI papers in academia, the harm and impact they cause far exceed the creation itself.

Now that the trend of AI development is surging, while embracing AI technology, please don't turn the Internet into a huge information garbage dump.

Reference links:

https://x.com/venturetwins/status/1939535063882027032

https://x.com/techhalla/status/1938905997038227921

https://x.com/venturetwins/status/1939889476467400986

https://www.nytimes.com/2024/03/29/opinion/ai - internet - x - youtube.html