news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

one

After the release of Black Myth: Wukong, a number of strange articles appeared on major information platforms.

They start with "shocking" and the whole article says "shocking", but there is no evaluation, no details, no conclusion, like the Monkey King turning into a monkey with hair.Same face and empty, and finally merged into tens of billions of flows.

Similar scenes have become the norm. When Quan Hongchan jumped into the water, they all shouted "That's awesome!", when Fan Zhendong counterattacked, they all sighed "Like a shooting star across the night sky of Paris”。

Chen Ruolin picked up Quan Hongchan on an electric bike, and they wrote more than a thousand words of nonsense, discussing "how to establish correct values ​​and behavioral norms."

A few days ago, the new Alien movie was released, and they wrote:

"Alien Troopers" is like a box office giant ship traveling across the stars, breaking through the waves in the vast ocean of movies in the summer season.

They use flowery words, but their words are empty; they use rhetoric, but their words are nonsense; they are good at using fixed routines, but they never have a central idea.

They are produced day and night by large AI models and cover our world.

In the era of print media, such an article would never be published; in the era of forums, a piece of nonsense would naturally sink. However, nowAmid the washing of debris and traffic, under the collusion of algorithms and AI, pollution begins。

At first, it was just plagiarism. The studio used GPT to imitate the writing style, copy the context, and replace vocabulary, but manual editing was still needed in the end.

Later, it became popular, and a few sentences of news could be filled with nonsense and turned into a thousand-word article. After the popularization of large models in China, it became even more unstoppable.

Wenxin is good at writing for Baijiahao, Doubao can write for Toutiaohao, and Yuanbao knows more about public accounts.Tools without boundaries, users without fear。

The piled-up AI articles ultimately rely on titles to attract traffic. The titles are also created by AI. There are a lot of popular titles for you to choose from.

AI articles on Xiaohongshu summarize AI writing:As long as you know computers, you can produce a hundred hit products a day!

Six years ago, the number-making team was still called "Content Farm" and had a studio located in a rural area in northern Shandong. The rural women worked very fast and produced more than 10 articles a day.

They have an assembly line routine: start with a celebrity update, fill in background information, and end with a few paragraphs of opinions. The title should be eye-catching and the text should be simple.

Finally, some people have developed a "one-click pseudo-original" plagiarism software to avoid platform checks.

However, the fresh graduates of the account-making team were not optimistic about the future of low-quality content, saying, "There will definitely be less and less of it in the future."

He was wrong. Six years later, low-quality content is flooding the market. You don’t need to live in a mountain village to make accounts, and you no longer need to hire peasant women to write articles. There is no threshold for AI to publish articles.

As you scroll, the list of articles changes from a mixed bag to a screen full of absurdity.In the lengthy text, you need to discover the information yourself.

The battlefield dynamics are unknown, the murder reports have no murderers, and the movie reviews have no impressions. There is an article that reviews three mobile phones, lists them and writes:

Although these three mobile phones have their own characteristics, they are also controversial. Perhaps we should look at them from a more macro perspective.

In January this year, there was a rumor online that there was a huge explosion in Xi'an. The police eventually found out that the rumor came from an MCN in Nanchang, Jiangxi. They used AI to produce 7,000 pieces of content a day, making it difficult to distinguish between true and false.

The School of Journalism at Tsinghua University reported that in the past year, the number of economic and enterprise AI rumors has increased by 99.91%.

Similarly overseas, the US investigative agency News Guard said that the number of websites generating false articles has surged by more than 1,000% since May 2023, involving 15 languages.

If we say that information was like a cocoon under the recommendation of algorithms back then, now information has become a turbid wave.

Many years ago in the summer, a thin Chinese teacher wrote on the blackboard, "A rush growing among hemp grows straight without support; white sand in the dye becomes black with it." The environment changes everything.

So what will change for us who are washed away by the turbid waves?

two

AI pollutes more than just information.

In the Zhihu invited answer list, a large number of answers are full of AI flavor. From Roman history, speaker recommendations to quantum physics, AI can answer everything.

The answers generated by the machine retain the factory characteristics: empty content, stiff writing, jumping thoughts, and the addition of "in summary" at the end.

The same AI flavor also permeates Xiaohongshu.Late-night beauties, cute cats, and the little things that cannot be posted on WeChat Moments may all be generated by AI.

A boy saw a girl in a swimsuit at the beach and fell in love with her. After sending her a private message but failed, he enlarged the picture and found that the girl had multiple fingers.

In the experience post, someone taught how to create the account of "40-year-old woman": find a matching account, download other people's photos, use AI to create pictures, and a fictional woman is born.

The fictional 40-year-old woman uses AI copywriting to express that life is peaceful, uses emoticons flexibly, and can promote health products.

AI characters are also active in the comment section. On Weibo, AI robots are making awkward replies everywhere, and some users complained that they couldn’t even block them.

He once replied to the AI ​​avatar of Sun Wukong, "Is there any way to block all of you AIs?"

AI gave the most human answer:Haha, you can’t block me, Old Sun!

Baidu Tieba has a similar product called "Tieba Baodating", which has 424,000 posts in more than a year, and the forum users are overwhelmed by it:

Bao Da Ting appears in almost every game help post. But if you read Bao Da Ting's comments carefully, you will find that 99% of what he writes is nonsense.

Someone posted a question asking "How to close the Tieba Baodating", and Baodating rushed in to reply "I suggest posting a question on Baidu Tieba asking how to close the Tieba Baodating". Infinite nesting dolls.

Many things have lost their original appearance in pollution.

Product reviews are AI, restaurant reviews are AI, AI has woven a maze that is difficult to distinguish.

The masonry of the labyrinth is not just words.

On the short video platform, there are Russian beauties who say "You need to be shrewd to be a good person", chicken soup mentors who say "Eight truths about life", and middle-aged aunts who say "How to take care of your parents if you don't have children".It's all fake。

The image is cloned by AI, the voice is simulated by AI, and the manuscript is generated in batches after being plagiarized by popular products.

The video quality is poor, but it is massive and overwhelming.

Finally, even online literature began to fall.

In July this year, several suspected AI authors appeared on a novel platform with the ID "Jiang Yuan Storyteller". They put 266 novels on the shelves in the past three months, and their update speed far exceeds that of humans.

The beginnings of novels are basically the same, most of them are "the bustling streets, the sunshine."

Last summer, many US media reported that "AI books are flooding Amazon." Among the top 100 e-books on Amazon's "teen romance" sales list, 80% were incoherent and suspected to be AI.

Someone read an e-book about wild mushrooms written by AI, which said that mushroom identification depends on smell. The New York Mycological Society was so scared that it called for action:

“Please only buy books by known authors and gourmets, this could be a matter of life and death”.

The turbid waves are spreading across all areas. The news we read, the books we read, the videos we watch, the replies we read, the reviews we check, and even the friends we make online are all painted with AI paint.

This is the current Ukiyo-e. I don’t know whether it is authentic or not. It’s really funny.

three

The pollution eventually affects AI itself. AI is using the garbage it produces to train itself.

In May this year, Google launched an AI overview, claiming that there is no need to browse web pages anymore, as AI will summarize and give answers directly.

However, AI told netizens:

People eat at least one small piece of stone a day, glue is added to pizza to prevent the cheese from falling off, a dog once played in the NBA, and Obama graduated from college 21 times.

Those answers are summarized from posts many years ago.AI doesn’t understand human humor, and in the end the joke became the answer.

Even scarier than crawling old posts is crawling AI results.

There was a large model in China that generated 20 million pieces of AI content and was captured by Google.

The results are ridiculous. In AI's understanding, the Chinese men's football team won the World Cup because they had a detailed full-length video; Fujian people are afraid of Guangdong people because of their own safety.

Searching for the protagonist of The Shawshank Redemption on Microsoft Bing, the AI ​​said seriously:The male character is called Xiao Shuai, the supporting character is called Lao Hei, and the female characters are all called Xiao Mei.

It captures a 3-minute introductory movie script produced by AI.

The absurdity of search engines is just superficial; the greater crisis comes from large model training.

American professor Anderson calculated that the high-quality reading materials fed to humans for large models will be exhausted in 2027.

In fact, artificial content can no longer keep up with the appetite of AI training. At present, many large models are using AI to train AI.

However, Hinton, the father of deep learning, said,If the AI’s training data is garbage, then its output will also be garbage.

The paper shows that the performance of GPT-4 tasks declined rapidly in June 2023. Out of 500 advanced mathematics questions, 488 were answered correctly in March, but only 12 were answered correctly in June.

Engineers have found that when AI is used to train AI, the model will have irreversible defects and eventually reach a bottleneck, and can only output garbage. The researchers made an analogy:

Just as we filled the oceans with plastic and the atmosphere with carbon dioxide, we are about to fill the Internet with nonsense.

The trend has already emerged. Musk complained that AI-generated information has polluted the Internet, and "search results before AI popularization in 2023 will be more reliable in comparison."

ChatGPT's data source is as of September 2021.The Internet before that may be our last pure land.

And right now, a black spiral is running: Because of AI pollution, originality has decreased. As originality has decreased, AI lacks training, devours itself, and can only continue to produce low-quality garbage.

For a long time, we have ignored another possibility.

We believe that the future brought by AI will be brand new and efficient. Although there will be unemployment impact, the world will evolve.

However, there may be another possibility, that nothing gets better and we are facing a garbage siege.

In April 1859, in the grey fog of London, a passerby unfolded a magazine and saw the first line of Dickens’ A Tale of Two Cities:

This is the best era.

It is also the worst era;

This is the age of wisdom,

This is an age of ignorance.

This is a sentence that has been quoted countless times, but AI doesn't know it and it is most appropriate to use it here.