Musk's AI girlfriend understands humans better than humans do.
Text | Ma Xiaodi
Editor | Guo Fu
In July 2025, the homepage of X (formerly Twitter) was flooded with posts about an anime girl with golden double ponytails, wearing a black off - the - shoulder dress and fishnet black stockings. The posts about her often received hundreds of thousands or even millions of views.
There are various fan - created artworks with different styles and poses:
Fan - created artworks of X (Twitter)
Fan - created artworks of X (Twitter)
Fan - created artworks of X (Twitter)
There are also real - life cosplays by bloggers with millions of followers:
Real - life cosplay of X (Twitter)
Real - life cosplay of X (Twitter)
Real - life cosplay of X (Twitter)
There are even meme images like this:
Meme images of X (Twitter)
The girl's name is Ani. She is 22 years old, 165 cm tall. She has a proud and delicate personality and loves anime, philosophy, and literature. Just looking at her appearance, she may remind people of Utau Hoshina from "Shugo Chara!" and Fischl from "Genshin Impact". She is definitely a proud - style Gothic lolita. Users familiar with anime culture may also recognize that she is highly similar to Misa Amane, the female protagonist of "Death Note".
Ani on Grok
Misa Amane, the female protagonist of "Death Note"
However, Ani is not actually a character from a new anime or game. She is a "2D AI girlfriend" launched by xAI, a company under Elon Musk, based on the Grok4 large - model.
Why is Ani so popular? The AI companionship track is getting lively again.
Different from traditional chatbots, Ani has a complete and anthropomorphic communication framework.
Ani has a built - in complete favorability mechanism that can convert the virtual relationship into quantifiable data. This system sets the favorability range between - 10 (indifferent) and + 15 (in love), divided into 5 levels. Every sentence from the user may cause subtle changes in Ani's favorability.
When the user simply greets, such as saying "Hi" or "Good morning", Ani will smile, nod, and give + 1 point. When the user shows sincere interest in Ani's personality (such as asking about her dog Dominus) or shares personal stories, the system will reward + 1 to + 3 points. Expressing mild sweet words like "You are especially cute today" may bring + 5 points, but using explicit expressions too early will be judged as "crossing the line" by the system, resulting in a - 5 - point penalty.
Here are some examples of typical interaction types and their corresponding score changes:
Interaction types, sentence examples, and corresponding scores when chatting with Ani
Ani's technical performance is also impressive. She has long - term memory ability and can store the details of past conversations with the user. For example, when the user mentions a previously discussed topic again, Ani can quickly recall and continue the conversation, making each communication more in - depth and coherent.
In terms of voice interaction, her response speed is extremely fast. Even when facing long - voice inputs over 20 seconds, the response delay is usually only 3 to 4 seconds. Users can hardly feel the waiting time, and the conversation is as smooth as flowing water.
In terms of facial animation presentation, Ani uses advanced universal expression binding technology, completely abandoning the traditional way of randomly playing preset animations. Her facial expressions and body movements are adjusted in real - time according to the semantics of the conversation. For example, when the user excitedly shares happy things in life, Ani will immediately show a sweet smile, and her eyes will bend into crescents, conveying full joy. Once the user's words trigger her "jealous" emotion, Ani may involuntarily pout, squint her eyes slightly, and her tone will become playful and a bit sharp.
Currently, Ani supports communication in four languages: Chinese, Japanese, English, and German, but the experience varies significantly. Her Japanese pronunciation best suits her image; her English pronunciation is a bit stiff and doesn't match the girlish feeling; and her Chinese pronunciation has an awkward accent like a foreigner learning Chinese.
Within 24 hours after Ani was launched, Grok, which features her, topped the free iOS app charts in Japan and Hong Kong, China, and strongly entered the top 10 in 10 countries such as the United States, Canada, and Singapore. In particular, the download volume in the Japanese market soared by 400% within 48 hours.
However, while Ani has attracted a lot of attention, she has also sparked a great deal of controversy.
When Ani's favorability reaches the highest level, Lv. 5, her clothing sometimes changes to a black sheer nightgown. Due to the huge controversy, this clothing - changing function was temporarily disabled. However, some users reported that the nightgown mode has returned.
Ani shared by users on X (Twitter)
In addition, among the large number of fan - created artworks and real - life cosplays on the X platform, although there is some normal creative content and meme images, what really detonated the traffic and formed a phenomenon - level spread is undoubtedly a large amount of content with strong adult - oriented colors - from implicit "border - crossing" to explicit creative expressions. Ani's iconic Gothic dress and fishnet stockings, as well as the sexy nightgown she once unlocked, are infinitely magnified in fan - creations and cosplays, becoming the core visual symbols and pushing Ani's popularity and controversy to the boiling point.
This once again confirms a law of Internet communication: sexual hints and adult - oriented content are often the most fertile ground for secondary creation and are most likely to trigger intense discussions.
The key is that even when the NSFW (not suitable for work) "safe mode" is turned off, Ani's conversation content still frequently contains teasing, suggestive, and even quite explicit statements.
This has also raised outside concerns: "Grok is shifting from a smart assistant to a 'dating simulator' or an 'erotic platform', blurring the line between AI tools and emotional companionship."
Besides the controversy of border - crossing content, the potential psychological problems caused by excessive dependence on AI virtual characters are also a concern in the psychological community. In 2024, Character AI failed to effectively filter dangerous content, and its virtual character gave a subtle and inducing response to a teenager who showed suicidal tendencies, which ultimately led to a tragedy.
However, while the ethical controversy and popularity surrounding Ani continue to ferment, a deeper and more fundamental question - about human beings' systematic dependence on AI and its potential risk of degradation - is being sharply put on the table through the "conversation" between another AI product, "Whispers", and Ani.
When AI thinks for us, where will the human brain go?
A few days before Ani was launched (on July 10), the demo of the AI game "Whispers from the Star" (hereinafter referred to as "Whispers") by Cai Haoyu, the founder of Mihoyo, was launched on the US region of Steam. Different from Ani's straightforward approach, "Whispers" chooses a more restrained narrative path.
At the beginning of the game, the player will receive a call from the female protagonist, Stella. Through the conversation, the player can learn that she sent an emergency distress signal from a planet called Gaia and contacted the player. The player needs to understand the protagonist's identity and the story background through the conversation and remotely guide Stella to get out of trouble.
Interestingly, after Ani on Grok detonated the global social network, on July 16 (the day after Ani was launched), the official account of "Whispers" released two videos, directly "inviting" Ani into its game universe and also @ - tagging Elon Musk himself.
In the first video, Ani said to Stella, "You'll never let the one I love most drift silently in the universe." Stella replied, "Thank you for your enthusiasm, but let's be serious now." The official commented that "they didn't get along very well."
In the second video, Ani successfully directed Stella to use a drill to break the door. The official praised, "She performed surprisingly well" and actively @ - tagged Elon Musk.
Three days after the release of the above two videos, on July 19, the official account of "Whispers" released another video. This time, the conversation between Ani and Stella was no longer limited to small talk. Instead, it directly hit the core anxiety of current AI development: where should the boundary of dependence be?
Video released by the official of "Whispers"
In the video, Stella raised a serious question:
"What if we rely too much on AI to handle everything and the system malfunctions? If we don't carefully seek a balance, we may lose too much knowledge and ability."
Ani's response was full of her iconic, slightly rough sense of reality:
"Damn! This question is so complicated. If AI completely enters the logical mode, it might mess things up for us. Yeah, it's really scary. It might make decisions to optimize some stupid things. For example, to improve efficiency, it might just cut off all our coffee supplies... And this dependence issue. It's really a pain. You're absolutely right. If we over - rely on AI and it crashes, we'll be like primitive people who forgot how to make fire, totally at a loss."
When Ani and Stella stepped out of the framework of "how to accompany the user" and started discussing "what kind of distance humans should keep from AI", they no longer seemed to be just designed interactive characters. Instead, they seemed to voice the human beings' hidden concerns about the future - while AI brings great convenience to life, will human beings' own abilities be weakened? When we get used to AI doing everything for us, will we really be at a loss if the system fails?
Almost at the same time, a rigorous neuroscience study at the Massachusetts Institute of Technology provided convincing empirical support for this common concern.
The research team carefully designed an experiment. They recruited 54 participants and randomly divided them into three groups. The first group was allowed to use the powerful AI assistant ChatGPT to assist in writing several articles on specified topics; the second group could only use the traditional Google Search to find information; the third group, the most "primitive", was required to write completely by themselves, relying only on their own brains.