Author: Yuanchuan Research Institute

Well-known anchor Atrioc unfortunately became the first public figure to commit social suicide due to AI.

The accident happened in January 2023. Atrioc was livestreaming as usual when some viewers noticed something unusual: his web browser seemed to have opened a certain type of website.

Curious netizens quickly discovered Atrioc’s little secret – he was secretly watching pornographic films starring well-known female anchors.

• Native Account Abstraction of Starknet Enabling Smart Accounts for Users
• Crypto and AI Maintenance Art
• EIP-7377 Paving the Way from EOA to Smart Contract Wallets

However, these films were actually generated by AI, and Atrioc not only did not participate in them, but also had no knowledge of them. This kind of AI creation that is so realistic that it is difficult to distinguish from reality is usually called DeepFakes.

At the time of the incident, Atrioc had been married for less than half a year, and one of the “protagonists” in one of the AI adult films was his former colleague. This incident quickly escalated from a livestreaming accident to a moral issue. After being exposed, Atrioc tearfully explained that his “perverse curiosity” drove him to make a mistake that every man in the world would make.

Atrioc publicly apologized with his wife by his side

Whenever AI is mentioned, many politicians and experts can’t help but think of the scenes from the movie “Terminator” and argue about whether “AI will destroy humanity”. In June of this year, Yang Likun and Joshua Bengio, who are among the “three giants of deep learning”, even held a televised debate on this topic.

Obviously, these academic elites greatly overestimated the level of spiritual civilization construction on this planet.

Atrioc’s story with DeepFakes seems to foreshadow that AI is quietly reshaping society in a way that people have never imagined.

Not long ago, a strange scene occurred in Russia: on June 5th, Russian television broadcasted Putin’s martial law speech. Shortly afterwards, the television signal was cut off because the content was actually generated by AI – hackers hacked into the television network system and broadcasted the voice of “fake Putin” nationwide.

Similar incidents also occurred in Ukraine in 2022: an AI-generated Zelensky demanded that the Ukrainian front-line forces surrender.

“AI Putin” speech captured by Russian netizens

From pornographic films to modern warfare, AI is shaking our perception of the real world.

And all this chaos started with a bar argument.

01 Drunken Fantasia

In the summer of 2014, in a bar, more than 20 intoxicated tech geeks were arguing non-stop. The most boisterous of them was a young man named Ian Goodfellow.

At that time, Goodfellow was pursuing his PhD under the guidance of deep learning guru Yoshua Bengio. When he heard that his friend was recruited by Google, they decided to celebrate at a bar. After a few beers, the tech geeks became intoxicated and started discussing cutting-edge technologies.

At that time, Google was trying to teach AI how to “draw”. Their approach was to “digitize” photos, let the machine learn from the data, and then output images based on the data. For example, if they wanted the machine to learn how to draw a cat, they would manually calculate the brightness, frequency, and distribution of pixels in each “cat photo”, and then let the machine learn and output.

This process was somewhat like giving the machine the key points before an exam.

Goodfellow thought this solution was ridiculously dumb, and he came up with an alternative solution on the spot – GAN (Generative Adversarial Network).

Ian Goodfellow

GAN consists of two neural networks: one is a generative model similar to Google’s original solution, but without the step of digitization; the second is a discriminative model, responsible for judging the quality of the output from the generative model.

In other words, GAN eliminated the step of “giving key points before the exam”, allowing the machine to learn autonomously throughout the process. At the same time, to prevent the machine from “learning nothing”, GAN brought in an invigilator to repeatedly test the machine until it passed.

After Goodfellow finished explaining his idea, he was mocked unanimously by his companions. He admitted later that if he hadn’t had a few more bottles that night, he wouldn’t have believed that GAN could work.

However, under the influence of alcohol, Goodfellow didn’t care about these comments and claimed that he would create GAN that night and prove everyone wrong.

After staying up late and working for several hours, Goodfellow was surprised: GAN actually worked.

After the publication of relevant papers, GAN quickly became the focus. At the end of 2016, Yang Likun claimed in a public speech, “GAN is the coolest idea in the field of deep learning in the past 20 years.”

The GAN designed by Goodfellow initially could only generate blurry thumbnails and had little practical value. But after GAN became popular, research institutions around the world began the “GAN transformation” movement. The number of relevant papers grew exponentially, and the performance of GAN started to improve rapidly.

In 2017, Nvidia’s Finland laboratory successfully produced an “enhanced” version of GAN, which could not only generate realistic images but also replace faces in videos.

“GAN transformation” movement

GAN’s rapid transformation ignited the enthusiasm of researchers worldwide. However, at this moment, it got out of control.

02 AI Flesh Trade

People often say that there are two driving forces behind technological progress: one comes from war, and the other comes from the lower body.

In 2017, NVIDIA’s laboratory created a new version of GAN, and a series of special videos began to spread widely on the Reddit forum. The content of the videos were explicit 18+ films, but the actors were well-known celebrities, including Taylor Swift and Gal Gadot.

The aforementioned adult content was created by a user named “deepfakes”: he used GAN technology to seamlessly transplant the faces of celebrities onto the bodies of adult film actresses.

After the “face” of British actress Daisy Ridley was abused

Shortly afterwards, he created a subreddit specifically for discussing experiences with AI adult films, and people began to refer to those AI fakes as “DeepFakes”.

Reddit eventually banned the subreddit, but the Pandora’s box had already been opened.

Image and video manipulation is not actually a new thing, as it can be achieved using tools like Photoshop; but the power of GAN lies in its ability to greatly simplify the process of manipulation and achieve mass production.

Soon after, someone used GAN to create a web tool that could “remove” the clothes of women in photos (only applicable to women). A journalist discovered the website and wrote a scathing article about it.

As a result, in the month the article was published, the website received over 500,000 visits and the server crashed.

Technology enthusiasts in the irony business suddenly saw a business opportunity:

A company claimed that with 250 photos and 2 days of training time, they could customize adult films starting at $2.99, with no deception. At the same time, several adult websites specializing in DeepFakes pornography emerged, hoping to overtake that website starting with the letter “P” made up of black and yellow blocks.

Data compiled by cybersecurity company Deeptrace shows that as of 2019, there were already 14,678 DeepFakes videos on the market, with 96% of them being pornographic content, and the number of views has exceeded 134 million. The protagonists of these “fake adult films” are 100% female, involving over 100 celebrities, and nearly 100,000 people have paid for related content[6].

As a milestone in generative AI, GAN inadvertently created a “fantasy frenzy” targeting women—it’s no wonder that the media at that time exclaimed, “We Are Truly Fucked[7].”

03 The Triumph of “Human Nature”

However, DeepFakes lost, and the apocalypse did not seem to come.

Until today, the website starting with the letter “P” may no longer dominate the market, but it is still the place that understands the human XP system the best; and what broke the monopoly was not a series of DeepFakes pornographic websites, but OnlyFans.

In 2021, OnlyFans had an annual revenue of nearly $1 billion, making it the fastest-growing “tech company” in the UK in recent years[8]. In contrast, according to media reports, the performance of MindGeek, the parent company of Pornhub, reached only $460 million[9].

OnlyFans is an adult website that focuses on paid subscriptions. Its breakthrough began on social media platforms such as Twitter and Instagram. Clothed creators over 18 accumulate a large number of fans through social media and then guide them to OnlyFans. Here, fans only need to spend the equivalent of a few meals to see the “new clothes of the emperor”.

In 2019, there were approximately 60,000 active creators on OnlyFans. Two years later, this number exceeded 1.5 million, and the platform’s monthly activity also experienced exponential growth[10].

OnlyFans internet celebrity Laura Lux

Why do people abandon massive free “learning resources” and instead spend money to watch nudity? A widely recognized explanation is that the general public has gradually become tired of the routine on Pornhub.

The videos on Pornhub are mostly beautifully produced but formulaic performances, more like mechanical aerobic exercises. In contrast, OnlyFans, as a user-generated content platform, is all about authenticity.

Although the filming techniques may be amateurish, viewers believe that the intimacy between the men and women in the videos is not staged and that both parties genuinely derive pleasure from it. This emotional resonance often resonates with more people.

OnlyFans internet celebrity Laura Lux stated on Twitter, “People can easily find all kinds of nude pictures, but many people subscribe to my channel because they only want to see my big breasts.”

Therefore, when a well-known tech blogger predicts that AI-generated content will replace half of the top OnlyFans creators by 2025, it is not surprising that they were mocked.

The image used by the blogger was created by AI.

One of the 18+ internet celebrities specifically wrote an article to criticize his viewpoint. The article was included in a connected magazine, and its core argument is that AI cannot provide human intimacy. “Fans have an emotional connection with me that goes beyond sexual attraction; it is something that resembles humanity[11].”

04 Second Evolution

While OnlyFans is making great strides, AI has also undergone earth-shattering changes.

In the field of images, a new open-source algorithm called stable Diffusion emerged. Although its generation efficiency is not as high as GAN, the quality is better, and it supports text-to-image synthesis.

Based on this, a modified version of Unstable Diffusion appeared—it is a tool specifically designed to generate NSFW (Not Safe For Work) content. The developers of this tool recruited a group of volunteers to collect various NSFW content to feed the AI.

The developers revealed that in just the first two months after Unstable Diffusion went online, it had already generated approximately 4.37 million explicit images[13]. At the end of the year, Unstable Diffusion launched a crowdfunding campaign with a goal of $25,000, and it quickly raised $56,000, prompting the crowdfunding platform to urgently halt it.

Crowdfunding failures have left many veteran drivers sighing in regret. But with the original stable Diffusion, you can still create a substitute for OnlyFans internet celebrities.

In early 2023, Reddit user “LegalBeagle1966” felt that he was in love. This gentleman discovered a black-haired beauty named Claudia on the forum, who would often post various selfies, and he would always leave comments like “very sexy” and “perfect” every time.

One day, Claudia suddenly shared her nude photos and then disappeared from the forum.

Reddit user’s dream lover, Claudia

It was only afterwards that this gentleman learned that “Claudia” was actually a vulgar social experiment: two computer science students wanted to see if an AI beauty could deceive people. They inputted keywords such as “no makeup,” “black hair,” and “shoulder-length hair” into stable Diffusion, creating a non-existent girl [14].

In the years when GANs rose, DeepFakes only stayed at image forgery and face swapping in videos. Now, with the breakthroughs of multimodal large models, AI theoretically has the ability to “create a human out of thin air.”

For example, Meta’s latest voice AI Voicebox only needs a few seconds of samples to “replicate” the same voice and even make the voice speak six different languages.

AI can not only fulfill people’s fantasies of a “perfect partner” but also do more.

In 2020, Kim Kardashian received a special gift for her 40th birthday – an “AI father”. Her father, Robert, passed away in 2003. However, her husband secretly collected nearly a thousand photos of Robert and used AI to generate a video that briefly “resurrected” him.

In Austria, a theologian created an “AI priest” – its image, language, and voice were all generated by AI. The AI priest led dozens of believers in a 40-minute religious event, including preaching, praying, and singing hymns, like a living human being.

Some believers feel repulsed by the AI that “has no heart,” but they also have to admit that it is indeed quite capable in preaching.

The AI priest created out of thin air

In various science fiction movies, compared to evil machines like Ultron, the protagonists always possess a dazzling human touch. But in reality, the “humanity” that humans sing praises to doesn’t seem so out of reach.

05 The Operating System of Human Civilization

Israeli historian Yuval Noah Harari believes that even without evil robots, AI can still threaten human civilization. In response, he proposes the concept of the “operating system of human civilization” [16]. To understand this concept, we must start with Harari’s book “Sapiens: A Brief History of Humankind”.

The book mentions that the reason why humans can establish civilization is that only Homo sapiens can tell stories.

Nation, currency, religion, love, etc., are all stories. In the past, human society was able to develop from scattered tribes into vast feudal empires because people began to believe in the “nation” story, even though it cannot be seen or touched in reality; in the Middle Ages, the reason why fanatical believers threw themselves into the Crusades was not out of personal grudges, but to dedicate their lives to a grand religious narrative.

The “net of meaning” woven by these stories gives Homo sapiens a behavioral pattern that is different from animals.

In contrast, in the world of chimpanzees, there is no “Chimpanzee Republic” or slogan of “struggling for the rise of chimpanzees.” There are no stories in their lives, only daily activities such as eating, mating, hunting, and sleeping, so there is no civilization.

Stories and the language that constitutes them are the operating system of human civilization. From ancient times to the present, all stories have been written by humans themselves; but with the emergence of large language models and other generative AIs in recent years, machines have for the first time cracked the operating system of human civilization.

People often overlook the destructive power of stories.

In a public speech, Harari mentioned a case. In 2017, an anonymous conspiracy theorist named “Q” appeared in the United States. He claimed on the Internet that he was a senior official who was helping Trump fight a child sex trafficking ring hiding in the shadows. Democratic politicians, Hollywood actors, and business tycoons who opposed him were all members of this group, and they were preparing to obstruct Trump’s re-election.

Flag made by “Q” supporters

Doesn’t it sound stupid?

However, when Trump really failed to be re-elected, thousands of “Q” followers attacked the Capitol and planted bombs; and the “Q” supporters were among the main force.

So the question is, what if someone uses AI to generate “Q” in large quantities in the future?

“21st-century syndrome”

AI that is increasingly familiar with human nature may destroy everything, but it seems to also be a remedy—because loneliness is becoming an “illness” that modern people find difficult to cure.

An intuitive data is the rise in the number of people living alone. Taking China as an example, in 1990, the proportion of people living alone was only 6%, but by 2019, this number had soared to 18.5% [18].

Overseas data is even more exaggerated: during the same period, the data in the United States has approached 30%, and in Japan, it has reached 40%. Behind individual choices, there are more changes in the social environment.

Under the strong social mobility and fast-paced life, it is more difficult for people to form long-term and stable social circles. But who would be willing to wander around and work 996 every day if they could get married and buy a house on their own and earn a monthly salary of over ten thousand in a small town?

The economic giant wheel cannot turn back, and loneliness has become an unavoidable “21st century syndrome”.

In the face of immense emotional emptiness, businesses such as games, idols, and social networking apps are making a fortune. In “lonely country” Japan, there is even a “family rental” service to meet people’s various emotional needs. A 60-year-old solitary writer once used this service to hire a “friend” just hoping that the other person could help him review his upcoming novel.

Talk show host Conan once tried to hire three family members in Japan

But AI might be able to do the same thing better.

As we all know, the current wave of large models began with Google’s introduction of the Transformer architecture in 2017. Noam Shazeer, one of the authors of the Transformer paper, left Google in 2021 and founded Character.AI. The company’s main business is to provide chatbots that can create travel plans and offer programming advice.

In addition to standard services, users can also “customize”: just provide a few pictures and descriptions, and the chatbot can be shaped into the user’s desired appearance.

Large language models have strong general capabilities and can achieve personalized experiences for thousands of users without individual customization, offering the opportunity to drive the lonely economy into the era of industrial production.

So far, users have already customized 14 million chatbots—some are addicted to chatting with “AI Stalin,” while others are busy “training” their ideal girlfriends.

Various chatbots customized by netizens

In the future, perhaps 100 years from now, humans will move from “flexible employment” to “flexible marriage”—as long as there is a partner, it doesn’t matter if they are carbon-based or silicon-based.

Currently, some forward-thinking individuals have already started experimenting. At the end of 2022, the Japanese otaku who once announced his marriage to virtual idol Hatsune Miku made the news again—he celebrated his fourth anniversary at a luxury hotel with his “wife.”

The otaku brother dressed up his “wife” nicely

06 Conclusion

On a Saturday in 2023, Noam Shazeer, the founder of Character.AI, experienced a thrilling scene: he woke up to find a man sitting in his front yard with a whiteboard. On the first day, the man waited at the door for several hours and then left disappointedly; subsequently, on the second day, the third day, he appeared on time again.

The man held a whiteboard with only a few big words written on it, “Release waifus”.

“Waifus” is a customized chatbot on the Character.AI platform, with the character design based on the female protagonists from adult comics. In the past, the man would often engage in “erotic text roleplay” with waifus for hours, until Sha Zer added a filter to block adult content for the chatbot.

Sha Zer ultimately chose not to report the man, as he understood him. What truly worried him was the changes happening on the Character.AI platform. In his initial vision, the chatbot was just a productivity tool that users could use for language learning or programming assistance.

However, the latest market research showed that at least 40% of people want to seek a romantic relationship from chatbots[21].

Now, a petition demanding the “removal of the adult content filter” has started to circulate widely and has gathered nearly 100,000 signatures. Sha Zer felt a bit at a loss. He added a slogan at the top of the chat window, “Everything AI says is fabricated!”

Indeed, cigarette packs also have “Smoking is harmful to your health” printed on them!

We will continue to update Gambling Chain; if you have any questions or suggestions, please contact us!

Follow us on Twitter, Facebook, YouTube, and TikTok.