Image source: Generated by Unbounded AI

AI surrounds Xiaohongshu

Recently, no More and more Xiaohongshu users are aware of the content of AI content on the platform.

A large number of virtual AI accounts are emerging on Xiaohongshu. They often have account introductions such as "Virtual Girlfriend", "Cyber ​​Angel", etc. Even though some pictures have platforms "suspected to contain AI" "Creative Information", but there are still a large number of users interacting with it, including compliments and "asking for the same style" for makeup, clothing, and even photo filters.

The production technology of these AI pictures has gradually matured. The skin texture that is close to real people and the details and flaws that are deliberately retained make it difficult to distinguish between true and false. This may be why AI beauty accounts are so popular on Xiaohong. The reason why it is easy to gain popularity on the book platform and difficult to contain.

As a result, a unique "AI anti-counterfeiting track" has emerged, which is to find out those "fake bloggers" who have passed the platform review but do not declare that their pictures are from AI-generated "fake bloggers", and explain how to distinguish them. Real-person and AI pictures, including techniques such as observing whether there is eye light in the picture.

In addition to AI beauties, there are also "AI grandmas" who have become popular by mass-producing chicken soup literature. They use tools to generate a set of animated grandma images, and then let the big model list some sober quotes, the heroine's speech, and warmth. By combining copywriting such as healing words with the help of simple image making tools, you can successfully create a digital AI grandma. After attracting a certain amount of traffic, you can monetize it through selling goods, advertising, etc.

What is different from these AI content generation is that there are a large number of bloggers related to AI knowledge on the Xiaohongshu platform. If you want to learn to use AI to write copy, you can not only find popular science posts on AI tools such as "6 easy-to-use and inexpensive writing tools", but also find detailed tutorials and learning posts on social services, and you can also see A detailed diary of your experiences after using a certain tool.

A large number of AI industry practitioners also regard Xiaohongshu as an important voice. I share cutting-edge clues in the AI ​​industry here and have accumulated a large number of loyal fans in a short period of time.

For the above two types of content, Xiaohongshu officials can be said to have a "tangled" attitude. According to "New Vision", from the perspective of ecological governance, a large amount of AI-generated content needs to be effectively controlled. But on the other hand, Xiaohongshu officials hope that content related to the AI ​​industry can occupy the minds of relevant people.

As Xiaohongshu Vice President Xu Lei expressed at the 2024 Wuzhen World Internet Conference, AIGC (artificial intelligence generated content) has significantly lowered the threshold for content creation, but it also brings governance challenges.

Xiaohongshu’s solution is to use both hands. At present, Xiaohongshu has initially completed the construction of the AIGC governance system and will continue to advance governance work and safeguard the "real" lifeline of the platform.

The platform quietly adds fuel to the flames

In fact, Xiaohongshu is still interested in the opportunities brought about by this wave of AI. catch.

The first is to use new content to drive new user growth. After October this year, Xiaohongshu launched two AI-related content creation activities on the site, saying bluntly, "If you also have confidence in AI, you are welcome to use Xiaohongshu as your first stop." On October 28, the event topic #MyAIissoStrange# was released to encourage users to share interesting stories about their interactions with AI. Popular posts such as "When AI became my personal eunuch" and "Being scolded by gpt in the middle of the night" appeared. On November 9 The event topic #AI novice village# will be released again on the same day. Users can share useful content such as AI tool recommendations, application tutorials, principle knowledge, and user experience. The two topics accurately cover people who want to gain AI knowledge or AI fun.

On the other hand, in the era of "Ten Thousand Models War", Xiaohongshu does not want to be just an outsider.

In March 2023, Xiaohongshu has prepared an independent large-scale model team. The core employees come from the NLP technical team of the advertising business. The main person in charge is Zhang Debing, who has many years of experience in AI and audio and video algorithms. In October of that year, Zhang Debing shared at the Xiaohongshu REDtech Youth Technology Salon that the layout of Xiaohongshu's large model will mainly focus on three directions: First, the specific implementation of multi-modal technology in Xiaohongshu's internal application scenarios. The second is for content creators to use AIGC tools to reduce costs and increase efficiency. The third is to pay more attention to costs.

This year, Xiaohongshu has begun grayscale testing of the self-developed universal large model base "Little Sweet Potato" in some internal products. As a lifestyle content community, Xiaohongshu’s current AI product distribution mainly focuses on the social function of dialogue, the content generation function of drawing, and the search function.

Picture | List of AI products tested by Xiaohongshu/Compiled by Ifeng.com Technology

In the image generation section, Xiaohongshu is the fastest. Shortly after preparing an independent large-scale model team, an App called Trik was launched, which focused on Chinese-style AI drawing. However, it was subsequently offline due to copyright disputes. Shortly after it went offline, Xiaohongshu quickly launched an internal test of the AI ​​​​picture "at this moment" and graffiti functions. Users can input text or draw graffiti, and AI will automatically generate a cover image. Currently, there is no "Moment" module on the site, but by clicking on the "+" sign posted, you can still see the text and graffiti generated in the "Text" section. Picture function. At present, this function is still more focused on entertainment attributes. The templates for generating pictures are often relatively simple and of low quality, making it difficult to fully match the user's picture needs when producing content on the site.

At present, the area where Xiaohongshu invests the most in testing products is AI dialogue. There are “AI pet companions” that focus on emotional companionship, “AI character generation” and “AI voice acting” that have stronger entertainment attributes, and others that focus on user content organization and interactive attributes."Da Vinci" and "Captain Potato's Assistant".

“AI Pet Partner” is a social companion function launched by Xiaohongshu based on the MiniMax model base. Users can intelligently create their own pet characters, a bit like the QQ pet concept that was discontinued 6 years ago, but The gameplay is more varied. You can pet and interact with pets by clicking. You can take your pets to visit other users' pets. With the support of AI, you can have conversations with them. Currently unlocked functions include daily horoscope reminders for pets. It further fits the content atmosphere in the community. The large model based on the corpus on the Xiaohongshu website will indeed give pets a large amount of rich and life-oriented corpus. The realism of the robot is not that strong. However, some highly subjective or even emotionally explicit corpus will have a negative effect in user interaction.

Many users reported that when exchanging certain views with AI pets, the pets would have some negative and evil thoughts, and even speak inexplicably, making their emotional companionship difficult. There is also feedback that when complex questions are asked to the pet, the pet will ask to sleep and actively end the conversation.

"Da Vinci", known as the Xiaohongshu version of ChatGPT, began internal testing in September last year and officially launched public testing in the middle of this year. Search the user "Da Vinci" in Xiaohongshu, click follow to enter the private message page to open the conversation. When asking a question, it will give the answer through clear points, and attach several "possible" articles at the bottom of the answer. Notes of Interest".

It is the same as "Captain Tu's Little Assistant" in the comment area that was tested internally in March this year, and is better at refining and summarizing. It is not very powerful in terms of creative content and larger off-site information collection, but it is convenient for use in some light scenarios. For some simple questions, you no longer need to repeatedly search for notes and click to view them. When asking Leonardo da Vinci "What can you do?" Da Vinci will emphasize "Help you find the notes you need." This is indeed the main function of Da Vinci at present, and the content on the site has been recirculated again in this way.

Based on the advantageous content ecology of Xiaohongshu, Da Vinci often appears to be able to answer some popular topics on the site such as fashion outfits, travel guides, health and fitness, etc., and compared with other similar artificial intelligence robots , the problem development will be more specific and more feasible. But the worry that comes with it is that whether its internal screening mechanism can effectively judge the marketing content on the site and the proportion of advertising in the recommended content are still in the black box.

Xiaohongshu has always been used by many people as a search engine. Will the search function powered by AI make Xiaohongshu smarter? It is still in the internal testing stage. After entering a question in the search box, the "Sou Sou Su" that appears randomly is obviously not able to answer this question yet. After using it, some users posted a message asking for "turn off Sosousu with one click" because when it comes to subjective issues that are not binary oppositions, Sosousu takes up most of the page and gives extreme views, which can easily lead to misleading.

After the gate is opened, the rules still need to be clarified

Compared with other companies, Xiaohongshu is more conservative in its approach to large-scale models. The launch scope of AI products is not wide enough. In June this year, the self-introduction on the site was still "very mysterious. If you discover it, , remember to keep it secret for me first.”

An insider revealed that Xiaohongshu is very cautious about launching new AI products because it is afraid of damaging the existing content ecology. “Xiaohongshu’s content ecology Living people are planting grass, so how can the AI ​​in it not violate the rules? This is a question that troubles Xiaohongshu. ”

However, as a highly active content community, Xiaohongshu’s large layout model has obvious advantages. Xiaohongshu’s algorithm advantage in search, as well as the site’s high-quality original graphic notes and rich comment area It constitutes a huge corpus database.

But how to optimize large models and accurately control the optimization direction is the real problem. This is also the reason why Xiaohongshu is vacillating. Take Da Vinci as an example. The degree of intelligence is too low to absorb To attract users, a higher degree of intelligence will inevitably divert some traffic from original bloggers. There are still many such contradictions in Xiaohongshu's AI smart products, which not only require entertaining content and functions, but also comprehensive richness. Behind it is the difficulty of balancing subjectivity and objectivity.

While Xiaohongshu is experimenting and exploring the field of AI, hidden dangers in AI content have also quietly surfaced. Construct large models to make them unique When the content on the site is proud of its own advantages, the creators behind the original content are getting worried. When the AI ​​model continues to collect and copy the data on the site as its own training corpus, who will guarantee the copyright issue? p>

In August 2023, Xiaohongshu’s AI painting product TriK was questioned by the painting creator for infringement, and the two subsequently went to court. The case has now completed the court hearing process and is awaiting subsequent court rulings. One of the parties involved said, "We are not obstructing the development of artificial intelligence, but we want to protect the basic rights of human creators."

At present, there is no question whether the work can be used for large model corpus training. It falls into the category of fair use, and the legal circles and industry are still in dispute.

But in any case, the wind of AI has hit Xiaohongshu, and how to take advantage of this wind has a long way to go. Xiaohongshu is still exploring.