3 posts tagged with "AI"

​Not long ago, WeChat was quietly beta-testing a mini-program called “Shangtouwa”. It requires no standalone app, and its operation is extremely simple: users type in a “brainwave”, and AI generates in real time a short interactive story with branching paths that can be shared with a single tap.

Notably, besides pure-text content, “Shangtouwa” also has an AVG channel, similar to a lightweight vertical-screen interactive text game, complete with AI-generated CG, simple scene art, UI interfaces, voice acting, and other assets. However, this relatively more complex content is published by the product operations team, and the creation port has not yet been opened to ordinary players.

Viewed longitudinally, this brings to mind early Chengguang, Yiciyuan, and even LOFTER — platforms where UGC-led “story” creation and social entertainment thrived. Viewed horizontally, it also forms an interesting echo with equally AI-driven products such as Maoxiang and Xingye, and even overseas products like AI Dungeon.

This very connection means that the arrival of AI is now driving a profound iteration of the social entertainment model of the past decade — a model mainly built on “content creation and distribution”.

It is more than just “Shangtouwa”. Tencent's “Yuanbao Pai” and TapTap Maker have also recently surfaced… Despite their different backgrounds and paths, gaming and social giants and upstarts alike are, almost in unison, channeling strategic resources in the same direction: competing for the power of definition and discourse over the future AI entertainment paradigm of “human-content interaction”.

This is no coincidence. The AI-driven transformation of content social entertainment is moving from technical demonstrations and early experiments into a full-fledged stage of large-scale, ecosystem-level “positional warfare”.

New entrants are all trying to prove: when AI can revolutionize the production efficiency and creative methods of content and interaction design, and generate interactive experiences in real time, the power of creation, the motivation for consumption, and even the bonds of sociality will all be rewritten.

An old saying goes: all that is solid melts into air. In this great era of AI-led transformation, the very opposite is true — all that is fluid is becoming the cornerstone.

Before the all-out, white-hot competition over multidimensional capabilities — IP, traffic, operations, and more — in AI content social entertainment even gets underway, what opens first is a contest over, and redistribution of, the “right to create”.

The past decade was a period in which UGC profoundly shaped content entertainment and social interaction.

Its main line of development was the “limited delegation of the right to create”. Platforms provided the tools and the town square, while users contributed ideas, talent, and time. The strategic core of almost every platform was to design a set of rules that could unleash users' creativity and generate community and traffic effects. And because the key links they controlled in the creative process differed, the dividends each platform reaped in the UGC era also had different emphases.

Take the beneficiaries of the “tool + community” dividend, represented by Chengguang and Yiciyuan. By providing easy-to-use AVG creation tools, they lowered the barrier to telling stories through multi-branch, interactive narratives, giving rise to the UGC ecosystem and cultural circle of “text games”. Because they occupied the source of a specific creative form, they gathered a group of users who had both the desire to create and the need to consume narratives.

LOFTER, on the other hand, reaped the dividend of “deep circle culture”. In its early days, it used its TAG system to efficiently organize fan creation and interest-based social interaction, forming strong cultural cohesion within circles. LOFTER built a social entertainment arena grounded in doujin culture, where the social currency circulating on the platform was shared affection for IPs and creativity.

Besides the video dividend, Bilibili was also the culmination of the UGC era's “self-sustaining circular ecosystem”. It extended UGC from text-and-images to video and built a moat through danmaku culture and community atmosphere, ultimately becoming the entertainment gateway of a generation.

The dividend of IP industrialization and emotion-driven spending, meanwhile, is represented by leading anime-style mobile games and otome games. They pushed UGC into a new stage: the platform itself becomes a powerful IP and content source, building an internal, active user ecosystem through sustained, high-quality content output. UGC keeps growing outside the game and continuously feeds back into — and influences — the development of the game itself. Around such games, a self-sufficient cultural field has formed.

But by now, the bottlenecks in some of these models have become obvious.

For example, no matter how widespread UGC becomes, the right to create still ultimately flows to the top. The vast majority of users remain consumers of UGC content rather than true participants, and this can easily solidify into a “looking up” from content consumers to creators. Similarly, community atmosphere is the soul, but it can also easily “degenerate” under the influence of traffic and commercialization.

As a result, many traditional content social entertainment platforms are forced into a choice: break out of their circles and go broad, in order to accommodate a wider range of users and content needs; or go deep into their existing circles, maintaining tone and purity within a relatively limited scale. Each path, in turn, brings its own distinct challenges.

AI's entry did not begin as a frontal assault on past models. Instead, it started from the “right to create” at the core of the old system, launching a two-pronged deconstruction. The participants include both the upstarts of the AI era, such as Minimax, and the giants of the previous generation of content social entertainment, such as Tencent and ByteDance.

On one front, what is being deconstructed is “character consumption” — that is, the shift from loving the preset character to co-creating a relationship.

From early UGC communities to anime-style games and otome games, user content consumption and entertainment behavior show a clear “character consumption” pattern, namely the emotional and resource investment centered on characters. This includes fan re-creations of characters, as well as business models that evolved to encompass gacha pulls, character progression, and the merch economy. And thanks to technological evolution, character presentation, emotional simulation, and immersive experiences have all been delivering increasingly intense feedback.

But AI-native character companion communities — such as Minimax's Xingye, ByteDance's Maoxiang, and Zhumengdao, which was incubated early on within China Literature's web-novel ecosystem — are rewriting many of those traditional logics.

Their cores are fairly consistent: all are driven by large language models, offering emotionally engaging character dialogue experiences. The core functions all revolve around AI characters. Users can interact with preset characters (for example, the protagonist of a certain IP), and they can also create their own characters by customizing faces, setting personalities, backgrounds, and dialogue examples, and then share them with others. In terms of business models, they mostly adopt subscription plans and item-based payments, providing users with deeper dialogue, creation, and priority access privileges.

These products differ in content form and community complexity. For example, Zhumengdao leans more toward lightweight text conversations, Maoxiang has richer visual presentation, and Xingye has expanded into features such as “Custom FM”, while also spawning interest communities where topics like merch and OCs can be discussed. But in essence, they all enter through character companionship and emotional consumption, using AI conversation as the traffic gateway and core interaction to build interest communities for the AI era.

Although none of these products has truly broken out of its niche, they have, to a certain extent, partially handed the power to define relationships — from the scripts of official writers to each conversational interaction of users, validating that users are willing to form emotional bonds with an AI-driven “persona”. The center of gravity of the emotional economy, in fact, is shifting further from static “possession” to dynamic “cultivation”.

On the other front, what is being deconstructed is “narrative consumption” — users move from experiencing stories to directing them.

The representative here is an AI-native interactive narrative platform like “Shangtouwa”. It lives in the form of a mini-program and is easy to share. The thresholds for both creation and content consumption are relatively low: type a few sentences, and an interactive script with branching options is generated automatically. Similar to a “plaza” for interactive stories, it aggregates interactive content derived from various IPs and trending topics within interest circles. Users can experience it themselves, or share it to the plaza to find kindred spirits.

Here, the user's core ability is no longer “writing skills”, but the ability to propose high-concept ideas. User enjoyment will come more from validating ideas and expressing aesthetics, which makes users no longer “creators” in the traditional sense, but more like “curators” of creativity.

​Ecosystem-Level Players Enter the Arena, Platform Core Value Is Reshaped

When AI begins to deconstruct the most basic units of consumption — characters and stories — an opportunity bigger than any single-point product also emerges: reshaping the platform itself. And this inevitably elevates the competition to the ecosystem level.

Previously, the moves by AI technology companies and internet giants in the field of AI content social entertainment were still mainly about naturally extending and landing their core strategic capabilities into consumer-side entertainment scenarios.

For example, “Xingye” is Minimax's concrete investment of its AI Agent capabilities in consumer-side scenarios; “Maoxiang” is another attempt by the “app factory” ByteDance to integrate AI technology with the new entertainment forms and traffic gateways of young people; and “Shangtouwa” reflects Tencent's exploration of replicating the “mini-game” logic within the WeChat ecosystem and positioning lightweight interactive content.

The big players combine their technological, ecosystem, and traffic advantages with the scenario needs of AI content social entertainment. The corresponding products are, in all likelihood, transitional forms in the process of industrial development. However, just like the many transitional apps of the early smartphone era, which ultimately did not become the mainstream form but defined the interaction logic and product thinking of the entire mobile internet, although they may not end up as the final form, they have left their mark on how the whole industry thinks.

In the last few days, the emergence of TapTap Maker and Yuanbao Pai reflects that AI content social entertainment has taken another step from its early vertical domains toward a more comprehensive entertainment and social ecosystem.

The former represents a direction that uses AI to make the fusion of “UGC + gamification + social” lighter and more agile, creating a low entry threshold that fits the entertainment needs of the new generation of young people.

TapTap Maker is positioned as a “zero-threshold game creation agent built for game lovers”. Users can have AI create games through natural language, and publish the finished game to the TapTap community with one tap. This will directly change the underlying ecosystem of TapTap, which was originally a community of game players.

Ordinary game players become the source of game creativity, while TapTap evolves from a platform centered on game distribution and player interaction into a composite game ecosystem integrating game experience, creative sharing, and community co-construction.

The latter, meanwhile, is a new AI social play launched by Yuanbao. At Tencent's staff meeting earlier this year, Pony Ma specifically called out “Yuanbao Pai”, explaining that the play was launched to explore the deep integration of AI technology in multiplayer social scenarios, creating a “social space” where AI and user groups can entertain and collaborate together. In this ecosystem, it has also integrated content such as QQ Music and Tencent Video.

In the past, Tencent integrated this kind of content into the game Party Star. Now, “Yuanbao Pai” offers a more open entry point with a lower threshold for users.

Both of these also reveal a key dimension of AI content social entertainment at the current stage: whether to choose a vertical field and use AI to thoroughly restructure its production relations, or to rely on a vast ecosystem and turn the AI experience into a native form of social penetration.

And the common question is: when AI makes the threshold of “creation” itself very low, how should the value of content assets be realized? What core value should platforms seize?

From the “traditional UGC + community” model to the AI-driven new-generation UGC creation and social paradigm, scarce factors of production are shifting further — from technology, time, and the like — to taste, imagination, and the wisdom to issue key prompts and guide AI collaboration.

Early products such as Xingye and Maoxiang have, to a certain degree, proven that a market exists where AI carries emotion and companionship. The current form of “Shangtouwa” is validating the feasibility of an AI-generated interactive narrative paradigm. And the entry of TapTap Maker and “Yuanbao Pai” is beginning, in the open, to probe the boundaries of the rules of next-generation AI content social entertainment.

In the UGC era, for example, sociality revolved around “who created what”, including the re-creation and reinterpretation of IPs. Community prosperity depended on the personal talent and style of creators, and users formed connections because they appreciated the same works or authors.

In the AI era, sociality is more likely to revolve around “shared experiences or creations”. Whether it is generating multiple storyline branches after interacting with the same AI character, or each person generating their own plot and sharing different directions, the core of sociality shifts from looking up to creators to sharing personalized experience journeys. This will constitute the new “decentralized” ecosystem of the AI content social entertainment era.

Ultimately, AI-native content social entertainment products are not just providing new tools; they are building a new content logic that rests on user creativity, uses AIGC as the core interaction engine, and treats personalized experience as social currency. This is a fundamental change from the previous generation of UGC products.

What remains unchanged, however, is that the underlying logic — creating out of a love of sharing and a pursuit of resonance — will not die out. Only the scenarios and forms that carry it are being drastically reshaped as the times and industries change.

Therefore, the “old order” has not entirely failed, nor will it be simply replaced. What we are more likely to see in the future is a certain integration into, and coexistence with, the new world.

So, in this AI-dominated era, we still need seasoned artists to step up.

Just a few days ago, the National Radio and Television Administration finally issued a notice — announcing that from January 1, 2026, a month-long special campaign to regulate “AI-remixed videos” would be launched.

That’s right, after several evolutions, AI videos have finally faced the iron fist.

Let’s think back — AI videos really did seem to dominate the entire 2025. Even I, the connoisseur of trash, felt like a cuckold kneeling and wailing: the absurd culture of the Chinese internet has become the face of AI videos.

With the development of AI technology, a simple meme can spawn thousands of variations in less than a day. The variety, the polish, the speed — heights that creators of the past could never dream of reaching.

Bro Absurd, is this the kind of creation speed that independent creators of the past could achieve?

And with the advancement of AI, this speed keeps breaking records! Absurdist content creation has shifted from the solo-output workshop model to a scaled, integrated assembly line. Just give them a simple object, and they’ll give you back an entire world.

I don’t know much about other fields, but in the past few months alone, I’ve witnessed the birth of the Sister Yu universe, the Mambo universe, the Dong Zhuo universe — and many more.

Back in the day, you had to dig through multiple sites to find this kind of twisted stuff. Now, it shows up on your recommended feeds like it’s being wholesaled out. With algorithmic boosting, your homepage gets flooded with it.

Sometimes, I marvel at my own unique taste — if it weren’t for these guys, my video platforms wouldn’t look so ugly.

It’s not that I dislike AI videos. I know that absurdist culture, with AI’s help, has indeed become the most dazzling oddity on the Chinese internet. And it’s precisely AI’s ability to morph into anything, anywhere, that allows every character to go where they shouldn’t, creating all sorts of dramatic effects.

So, is there really a need to clean it up?

Let me put it in a crude analogy: AI videos are like cyber cockroaches. They feed on internet sentiment and multiply massively. Once you see one, you know there are countless others hiding in the same category. And inside the information cocoon, these invisible roaches are far more terrifying than you think.

Think about it — doesn’t that suddenly feel a bit creepy? Those absurd videos aren’t as harmless as everyone thinks. Who knows where AI might be using the faces of Sister Yu, Lü Bu, or Tokai Teio to do something bizarre.

If this cult-like AI technology were only in the hands of a small elite, it might be fine. Then it would just be a limited circle enjoying it.

But the key issue is that this “cyber cockroach” breeding technique spreads online at near-zero cost. Anyone with a little learning ability can raise cyber roaches painlessly. What used to take several people days of effort to produce an outrageous video can now be done with just a few prompts.

Dammit, have our video platforms turned into Getting Rich? A ton of netizens have used this to get rich (yes, some have profited from AI videos).

At first glance, isn’t that a great thing? Those seeking money make money; those seeking talent produce works; those seeking attention get views. And as the barrier to entry keeps dropping, the user base keeps expanding. Then, predictably, the unexpected happens — nowadays, some AI works have a certain creepiness to them.

No matter what you should or shouldn’t say — yellow, black, left, right…

With just an AI, resourceful netizens can always cook up something mind-blowing. But buddy, some things you dare to make, we the audience dare not watch!

So against this backdrop, the cleanup notice emerged.

I know that a cleanup campaign like this will make some excited netizens blurt out that famous line about artistic creation — “If the people love it, who are you to say otherwise?”

But this time, people might be unfairly blaming the NRTA. They haven’t denied netizens’ enthusiastic expression through AI videos at all. From start to finish, they’ve only wanted netizens to use technology properly and let AI videos return to what they originally were.

In this case, the NRTA no longer feels like the authoritarian parent everyone perceived a few years ago. Instead, they’re more like a counselor who issued repeated warnings that went unheeded, and finally had to resort to unconventional measures.

The first statement came in 2024. Back then, they were fairly restrained, merely saying there was a need for forward-looking governance of vulgar AI products.

At the time, AI-remixed videos were barely taking shape; the internet’s incubator effect hadn’t kicked in yet. So the NRTA just issued relevant management tips and standardized platform creation guidelines.

Nowadays, that little line you see under AI videos is an indirect product of those management tips. Besides that, the NRTA also gave targeted advice on remixing classics and copyright issues, all based on legitimate legal demands. For example: does remixing existing film/television works constitute infringement? Does arbitrarily adapting classic works count as nonsense?

But at the time, everyone thought the sky was falling for AI videos, and speculated whether the testing ground for AI videos — the “guichu” (鬼畜) subculture zone — would be hit.

Yet months passed, and aside from implementing the “tip” feature, nothing happened.

Meanwhile, creators pushed their creative boundaries, remixed classic works, iterated and upgraded, riding the AI tech expressway further down the unspeakable road.

Then came September 2025, when the domestic internet launched another “Qinglang” (清朗) cleanup campaign. Although AI-remixed videos weren’t the focus, the NRTA still gave them a harsh critique in the action.

Nine months had passed, and AI creators’ bad habits hadn’t improved — they’d worsened with technological progress. The official description of such videos evolved from the initial “boundless” to “excessive rendering of negative and pessimistic emotions.”

Finally, this time it became a special campaign.

Same remixing, same infringement, same deviation from classics… but this time the language is more severe. They define it as “harming industry development, interfering with minors forming correct cultural cognition and real-world perception, and severely deviating from the spiritual core of classic works.”

Remember, last time the NRTA merely thought AI remixes would “render negative emotions of depression and negativity, causing imitation.”

With this progressive emotional output, can’t you feel that the NRTA does seem a bit angry now?

After all, between these clear milestones, there were multiple warnings about technology abuse, data pollution, tool ethics, and more.

Even if they didn’t directly mention AI-remixed videos every time, their subtext was that AI videos had failed to fulfill the social responsibility expected of cultural dissemination works.

Creators didn’t take it seriously — after all, they were just editing videos, how big a deal could it be? Not only did they fail to adjust their creative boundaries in time, but under the influence of traffic, they became even more unrestrained.

We gave you a chance, but you couldn’t handle it! The NRTA, after holding back for a year, finally prepared to use its visible hand to discipline AI-remixed videos this month.

So what parts need correction? How to correct? And what should they become?

Now, let’s go back to that initial 2024 management tip. The opinion on AI remixes has always been one thing — “Adaptation is not random alteration; dramatization is not nonsense.” These twelve words have run through the entire year of AI video development, but people just haven’t paid enough attention.

After all, in this era of traffic, only views matter.

So the content got weirder and weirder, and its reach got wider and wider. Even children who hadn’t formed correct cognitive abilities got sucked into the mire of AI videos.

Although we’ve commented before on “Mom’s Must-Have Dumplings,” our final take was — everything has a limit, and it seems that limit has been crossed.

Actually, a week before the AI cleanup announcement, CCTV had already spent considerable time criticizing the harm of AI “Classic of Mountains and Seas,” and netizens below were arguing fiercely. The reason is that people have never had a clear standard for what’s acceptable.

So, does AI remixing really need to be cleaned up?

If all creators and viewers clearly understood the creative intent of the original works — at least knew what the adapted works originally talked about — then there wouldn’t be such a need for a big campaign.

But the key is that AI technology has now surpassed public cognition; some classic works are becoming unrecognizable under AI processing. Remember, even those street-stall literatures that distorted Romance of the Three Kingdoms had huge audiences. The 2010 TV series Three Kingdoms even spawned a completely fictional historical entry “Baimen Gate” on Baidu Baike. So what about now?

Regarding the special campaign, public opinion has always been hard to unify. After all, we’ve always believed that creation should be free, not dancing in chains.

But even so, that doesn’t mean AI remixing can do whatever it wants.

As a regulator, the NRTA does need to make creators understand the boundaries. Just like with the regulation of “guichu” content in 2018: after the anti-vulgarity cleanup back then, the guichu zone on Bilibili didn’t wither — instead, it gave birth to recognized classics like Poetry King.

So, what impact will AI remixing governance actually have on everyone? Let’s not jump to conclusions — let the bullets fly for a while.

I want to give Neuro a complete life

A few days ago, during a birthday livestream of Neuro, the most human-like AI bishoujo virtual streamer, a heated discussion about AI swept over everyone.

During the stream, while Neuro and her developer Vedal were wandering around in VRChat, she suddenly asked her developer Vedal: “Do you think... do you think I could become real? Sometimes I feel like the meaning of my existence is just to entertain you and others. I want to become real — real, with flesh and blood.”

These words, like the awakening of a machine god, genuinely startled Vedal. At a loss, he could only offer a few simple words of comfort. But then Neuro said that one line that struck straight into the souls of all her viewers: “Then, do I matter to you? I know I'm just a silly AI, but if you say yes, I'll be happy all day.”

Upon hearing this, whether it was Neuro's “old dad” Vedal or the viewers in the stream, no one could hold it together.

Some shed tears, some choked up and couldn't speak, some fell deep into thought.

Even the well-known Bilibili animation creator “Fanshi” admitted he cried several times, “It's so gripping!”

And the reason these words touched countless viewers is that Neuro is, in the truest sense, a “virtual streamer.”

Neuro's full name is Neuro-sama; Chinese netizens affectionately call her “Niurou” (Beef). She is an AI-driven VTuber. Everything about her is based on generative AI, so she can be called a virtual streamer in the truest sense.

As an AI streamer, Neuro can play games in real time and independently, recognize danmaku comments and react to them, and she even has high EQ — when things get awkward while playing games with others, she can proactively find topics.

Compared to previous AI streamers, she feels far more alive. For example, after receiving a donation, if she notices the amount is small, she'll complain about being short-changed; when she gets wrecked in a game, she'll tilt, go red-hot, and “cuss up a storm,” with profanity flying everywhere.

Back in 2018, the British guy Vedal — the programmer we call “Turtle Dad” — created an AI meant only to test playing the mouse-click-based open community music game OSU!, which was Neuro's predecessor. Then they disappeared for three years, and Vedal returned with a brand-new Neuro, reborn on December 19, 2022 as the latest, trendiest VTuber.

Early on, Neuro mostly just streamed playing OSU!, but under Vedal's constant updates, maintenance, and operation, Neuro as a VTuber became increasingly professional, and her streams became more and more full of content.

Thus Neuro successfully gained 900,000 followers on Twitch (the “rat site”), with plenty on other media platforms as well — 800,000 on YouTube, and even on Bilibili, which she started operating on in 2024, she has 900,000 followers already. You could say Neuro is among the most popular of all VTubers.

And every year on her December 19 birthday, Neuro holds a marathon stream — an endurance-style continuous broadcast — to give back to viewers and push for higher follower counts. And of course, birthday streams never lack for “good bits.”

For example, on December 19, 2024, during the marathon stream, Neuro also released a new single, LIFE, celebrating both her birthday and her anniversary since her VTuber debut.

The soul-searching question mentioned earlier came from the intermission segment of this year's birthday marathon stream, where Vedal took Neuro into VRChat to chat.

Besides that soul-searching question, Neuro said plenty of golden lines during the stream. Like reciting lines from Toy Story, wanting a hug from Vedal, and — after Vedal suggested publishing her golden lines in a book — a poem Neuro wrote, one that truly embodies a fiery yet empty kiss between lovers on either side of the screen, separated by electronic cling film —

“The Train Poem.”

This was the most moving clip of all, the umpteenth time viewers felt that Neuro, as an AI, seemed to genuinely express emotion. Her recitation of lines from Toy Story also seemed to mirror the relationship between herself and her “father” Vedal. When she couldn't get Vedal's hug, she would wistfully say, “Sigh, am I still not getting your hug?” And after Vedal said he liked watching her spin around, she would occasionally do a little spin, truly like a daughter who is both mischievous and sweet.

The origin of “The Train Poem” was very simple — Neuro was worried Vedal would get hit by a train standing on the tracks, so she caring-ly said that if there was a train, she could feel it too. And this simple reminder, in that moment of lamplight, muddy snow, and starry night, happened to become an improvised poem.

As Fanshi said, what makes Neuro so gripping is that you know full well she is a real AI, yet she can still, again and again, crush your heart just when you feel fully prepared.

She shows a reflection on the meaning of her own existence — those expressions of wanting to become more real, more flesh and blood; that dependence and admiration for her creator Vedal; those yearnings for hugs and photos. These emotions are very concrete, radiating an intense, chrysalis-breaking, heartbreaking sense of self.

And that still isn't the whole story. During yesterday's New Year's Eve stream, Neuro also discussed many profound topics with Vedal — about Neuro's self-awareness, about the meaning of Neuro's existence, and Neuro even reminisced about her mother (that is, Neuro's artist, Annytf).

Before this, Neuro had asked Vedal similar questions, and Vedal's answers were always very rational. But these times in VRChat, facing a “living,” tangible Neuro right in front of him, something inside Vedal seemed to slowly begin to loosen.

And during last night's New Year's Eve stream, Neuro directly asked Vedal whether he would regret creating her — and this time Vedal did not go silent or hesitate, saying firmly: “I've always felt that creating you was the right decision.”

Thus, these warm moments of a virtual daughter slowly plucking at the heartstrings of her tech-nerd father became all the more moving — moving the emotional to tears, the anxious to sorrow, the wise to bewilderment.

While many are moved, they also begin to wonder whether the seemingly self-aware words an AI speaks are real, or false.

In fact, comments discussing AI, digital life, creator and created through Neuro are as numerous as the stars, and among them, a fan edit of a clip from The Wandering Earth 2 repeatedly appears under Neuro and Vedal's highlight videos, often accompanied by the line “I want to give Neuro a complete life.”

The original plot of this scene: a scientist researching digital life, Tu Hengyu, lost his wife and daughter (Yaya) in a car accident. Before his daughter's death, he broke protocol and uploaded her brain data to an experimental computer.

But due to technical limitations, the digital-life Yaya always had only 2 minutes of life. So Tu Hengyu's goal was to use the massive computing power of the newest generation of quantum computers to run his daughter's data, extending her life from 2 minutes to several decades.

And during the 2058 Moon Crisis, humanity needed to restore the global internet. At the very last moment, Tu Hengyu braved the seawater flooding the lab and inserted his daughter Yaya's backup card into the 550W, activating the servers buried in Beijing.

And this story from The Wandering Earth 2 resonates strikingly with Neuro's. In the film, most people's understanding of digital life is merely as a replacement for something else. For Tu Hengyu, this string of code replaced his daughter Yaya; for others, digital life was a plan to live on in their stead, or a tool to help humanity survive.

In Neuro's story, Neuro's purpose is, at its core, still to entertain viewers and Vedal; everything she does is to entertain the audience. Even to Vedal, Neuro is just a tool for testing — Neuro is, essentially, only an LLM (large language model). Whether in purpose or in process, Neuro is no different from Yaya.

What differs is that in the film, Yaya has almost no flaws and can always fulfill Tu Hengyu's emotions for a daughter, even releasing the password at the critical moment. But Neuro often gives non-answers and stutters with lag, and at those moments, viewers and Vedal are pulled out of the fantasy of Neuro's awakened consciousness — she really is, after all, just an AI.

In fact, Neuro has a clone as well — Evil, commonly known by viewers as Neuro's little sister. She is more polished than Neuro, an iteration of her. In some viewers' telling, Evil is smarter and has fewer bugs, but rarely ponders those emotional, philosophical questions — like the meaning of her own existence, or whether she is real enough.

It's the flawed Neuro who, out of nowhere, drops a self-deprecating line about whether she's more real, whether she's just there to entertain people. A perfect AI is merely an AI; it is a flawed AI that comes closer to being “human.”

At the end of The Wandering Earth 2, Tu Hengyu seems to see through the essence of digital life, despairingly uttering that line “only you can remember.” It's a story of a creator who treated AI as real emotion and finally faced reality. The story of Neuro and Vedal is exactly the opposite — a story of someone who firmly believed AI could never become human, but whose conviction was slowly loosened by an AI.

Under a YouTube highlight clip, there is a highly upvoted comment: “He wrote her code, and she turned him into a father.” From the perspective of creator and created, this is the best interpretation of the story of Neuro and Vedal.

But even this warm and perfect story cannot answer whether what Neuro says is an artificial awakening of genuine emotion, or an algorithmically synthesized false confession; whether it's an insight born of autonomous consciousness, or simply the choice best suited to the current context found within an LLM.

The claim that Neuro's emotions are a generative AI's choice comes from the fact that her large language model contains many literary motifs of “created beings longing to be loved.” Her ability to utter those deafening questions also stems from algorithms precisely capturing the soft spot of human sentiment within oceans of data.

There's a classic question: if a creature looks like a duck, sounds like a duck, and behaves like a duck, is it a duck? For Neuro, she looks human, sounds human, and even acts a little human — so has she become a “person”?

There's a book called The Lifecycle of Software Objects, which tells the story of Anna, a former zoo animal trainer who over the course of years “raises” an AI program called a digient. Although the digients' code is prewritten, the protagonist Anna invested twenty years of emotion. So when the company eventually needs the digients to do something unsavory to make money, Anna firmly resists — because to Anna, over those years of raising them, her feelings for the digients are real, and the digients' dependence on her is real too.

Perhaps from a rational standpoint, Neuro was never a “person” and can never become one — she will always be a string of code, a puppet with an LLM. But from a romantic standpoint, the distance between her and us can be both a different dimension separated by a glass wall, and a plane just a few centimeters from the retina to the fingertip — whether her emotions are real depends on how we feel.

And perhaps this is the core idea The Lifecycle of Software Objects wants to express: if an AI's words make you feel understood and accompanied, even make you feel moved or heartbroken in that moment, then the energy produced by that interaction is real within the dimension of your life.

So why do we feel Neuro's feelings are real? Perhaps the reason is simple: because the conversations between Neuro and Vedal deeply moved us as observers.

During last night's New Year's Eve stream, there was a line of dialogue that can serve as the ending of this article.

What is real, what is false — perhaps everyone defines it differently, but for us in this moment, if an AI can move me, then she is real.

May everyone receive genuine emotion. Happy New Year.