Source: Sina Technology (ID: techsina), author | Zhou Wenmeng

Image credit: Generated by Unbounded AI tools

Zhou Hongyi, chairman of 360, used one word to describe the popularity of ChatGPT, that is: Wanmo War. Up to now, at least more than 30 AI large-scale models have appeared in China, and the large-scale model competition has officially started.

Xiaohongshu will not miss this "train". Recently, it has been reported that Xiaohongshu has secretly prepared a "big model" team to participate in the research and development of large model technology. However, apart from the fact that the core staff of the large model come from the NLP technical team of the advertising business, and that several independent departments are exploring the direction of AIGC, more news is still in a "confidential state".

Xiaohongshu also did not respond to "whether it is engaged in large-scale model research and development". However, some insiders bluntly said that the large-scale model is not "multi-departmental participation" as rumored by the outside world, and many internal people lack confidence in this research and development. In the eyes of insiders, Xiaohongshu is a little "wavering" about the big model. On the one hand, it wants to make the content more diversified. On the other hand, it is also worried that "generative AI for content" has subverted the current quality of independent content. entry means.

As a representative enterprise of the "face value economy", Xiaohongshu has always had a deep sense of crisis for AI. Because the exquisite graphics and small knowledge videos based on poor information are the areas where AI is good at and easy to impact.

** Under stricter supervision, forced to enter the game? **

As of press time, Xiaohongshu has not made any response to the authenticity of the "Xiaohongshu layout model" reported by the outside world. However, combined with the "Trik" AI creation application previously launched by Xiaohongshu, and the attitude of the company's executives when talking about AIGC, it can be concluded that the large model of Xiaohongshu's layout and related product technology research and development are basically confirmed.

At the Xiaohongshu WILL Business Conference held in February this year, Xiaohongshu COO Conan said in an interview with the media that when talking about AIGC’s impact on future content forms, he pointed out, “With the advancement of technology and equipment, the content form will change. To become richer and more diverse, we should follow the technology and continue to pay attention to and promote the publishing tools we provide to creators.”

The high-level remarks, to a certain extent, represent that Xiaohongshu embraces the emergence of AIGC. As the large-scale model research and development and related layouts of leading Internet companies such as Ali, Baidu, Tencent, 360, and ByteDance continue to surface, a wave of AIGC entrepreneurship based on large models is constantly sweeping. In the eyes of industry insiders, "the current state of Xiaohongshu is not so much an active layout of large models, but rather a push by the external environment to have to layout."

It should be pointed out that as a social platform, Xiaohongshu’s technical capabilities are not outstanding. In view of the high computing power, data and labor costs of large-scale model development, the probability of Xiaohongshu’s participation in the development of basic large-scale models and industry large-scale models is low. High, and more likely to focus on large-scale model applications, focusing on user operations within the platform and the provision of creator tools.

The large model of Xiaohongshu, which "still holds a pipa and half hides its face", is mysterious, but it is also adding a little "ambiguity" to cover up.

On the one hand, with the drafting of the "Administrative Measures for Generative Artificial Intelligence Services (Draft for Comment)" (abbreviation: "Administrative Measures") and the start of soliciting opinions from the public, enterprises are in the process of promoting generative AI products and technologies. Faced with increasingly stringent audit requirements.

Xu Bin, an insider of the Chinese Academy of Social Sciences, revealed to Sina Finance that as the deadline for feedback on the management approach expires recently, “Although many companies have put forward many opinions and suggestions, the supervision is still more inclined to understand and generate from the content level. AI is not a knowledge production tool, and the future supervision trend may still be stricter.”

Take the cover of "AI Stefanie Sun" that has attracted much attention recently as an example. Xu Bin pointed out that "the emergence of a new round of generative AI based on large models, because before generating content, it must learn to imitate the character and other characteristic data of the object, this has actually jumped out of the infringer's 'unconscious' infringement of star portraits, knowledge It is not the scope of property rights, but as long as this technology is used, it will definitely infringe.”

This has led to the fact that due to the previous digital governance model, as long as there is a problem, the platform is basically responsible, or the platform bears joint and several liability. In the context of stricter regulation, Xiaohongshu, which has 260 million monthly active users, wants to participate in the large-scale model layout, and has to think about possible content governance and review management mechanisms from the beginning. And this, in fact, requires a lot of audit costs in addition to technology research and development.

**The key to success or failure depends on blogger satisfaction? **

On the other hand, for the layout of large models and even AIGC, whether it can win the support of bloggers on the existing platform of Xiaohongshu will also become the key to its development success or failure.

As an enterprise focusing on the beauty economy, the exquisite graphics and small knowledge videos based on poor information that Xiaohongshu relies on for survival are actually the most vulnerable areas to be impacted by AI. If Xiaohongshu rashly uses text notes, pictures, videos and other content on the platform for its own large-scale model development and training, but fails to sort out the content copyright and the corresponding blogger's interest relationship, this is likely to trigger platform blogging. Lords' resistance.

Before that, when other platforms were facing the trend of stopping updates, Lu Shihan, a technology Internet blogger with nearly 3.278 million followers, once talked about the "subversion of the content industry by generative AI" when talking with Sina Finance. is one of the main reasons.

According to Lu Shihan, "Because many bloggers who were good at text content creation did not have the ability to produce pictures and video clips, they recruited specialized personnel, which eventually led to a decline in revenue. 'The boss works for the employees', but as With the emergence of capabilities such as Vincent graphs and videos brought by generative AI, many self-media teams with low revenues have begun to consider reducing costs.”

After personally participating in many digital governance forums across the country, Xu Bin heard and witnessed a large number of real stories about the impact of generative AI on digital creators, and based on this, he summarized his views on generative AI—the emergence of generative AI , The subversion of the content industry is "tsunami-style", especially in the digital creation fields such as poster creation and cultural creation. Some digital creators with low local scarcity generally feel the impact of "crushing".

In addition to layout technology, in March this year, Xiaohongshu upgraded the live broadcast business from a secondary department under the Community Department to a new independent department. Live e-commerce and other businesses. This move is seen by the outside world as a turning point for Xiaohongshu to accelerate its embrace of commercialization.

However, with the arrival of this turning point, Xiaohongshu spent 9 years swinging between the contradictory state of "community atmosphere or commercialization". During this period, the founder team's operation strategy of "planting grass but not pulling weeds" led to the inability to form a closed loop of Xiaohongshu's e-commerce business. In the end, it could only "plant grass in Xiaohongshu and pull weeds on e-commerce platforms such as Taobao".

Today, Xiaohongshu’s grass-planting status and commercialization capabilities are constantly being questioned. Taobao, Pinduoduo, and JD.com have respectively launched grass-growing platforms such as Hangwang, Pinxiaoquan, and Planting Grass Show. Douyin, Kuaishou and even Zhihu have also begun to cut into the grass-growing track. This also indirectly caused the valuation of Xiaohongshu in the private equity market to drop by nearly half, from US$20 billion in 2021 to US$10 billion to US$16 billion in 2022.

The entanglement and swing of the internal strategy inevitably slowed down the commercialization of Xiaohongshu. Now, with the start of a new round of "big model" competition among domestic Internet platforms, the challenges of Xiaohongshu will only increase.