Source: Silicon-based Research Lab
Image Source: Generated by Wujie AI
Xiaohongshu (Little Red Book) is making another move in the layout of large models.
"Silicon-based Research Lab" observed that Xiaohongshu is currently testing its AI chat assistant "DAVINCI". Many users received the invitation for the internal test of Xiaohongshu's official account "Test Potato" as early as September this year, and had the opportunity to experience it by submitting a questionnaire.
According to the information previously released by "Xiaohongshu Official Recruitment," the AI intelligent assistant "DAVINCI" is one of Xiaohongshu's attempts in the field of AI. Earlier this year, Xiaohongshu launched an application called "Trik," which focuses on AI drawing. In July, Xiaohongshu launched graffiti and the "Now" function.
From the official announcement of the large model in the first half of the year to the transformation of business using large models in the second half of the year, major companies appear particularly anxious in this battle of large models. However, Xiaohongshu seems particularly low-key in comparison.
In terms of the layout of AI large models, the massive multimodal corpus in Xiaohongshu's content layer provides the data foundation, and Xiaohongshu's years of algorithm precipitation based on the search and recommendation scene is also one of its unique advantages.
However, as expressed by a user who has experienced "DAVINCI," there are concerns about Xiaohongshu's layout strategy for dialogue products: "Dialogue products are no longer new, but it's not clear what Xiaohongshu's layout strategy is."
What does Xiaohongshu's version of GPT look like?
"DAVINCI" is an AI assistant developed by Xiaohongshu based on a large language model, and its official positioning is "growing life artist." In the official information, Xiaohongshu describes the differentiated advantages of "DAVINCI" as follows:
(Users participating in the questionnaire have the opportunity to receive the internal test. Image source: Xiaohongshu screenshot)
"During the chat, DAVINCI will intimately understand the user's good questions. Perhaps the questions are very complex, but DAVINCI will provide personalized, efficient, and accurate detailed answers based on the summary and induction of numerous notes from the Xiaohongshu community."
Currently, "DAVINCI" is still in the internal testing phase and has not entered formal public testing. "Silicon-based Research Lab" learned from the experience of some users in the actual test that "DAVINCI" can basically meet daily Q&A and note summary needs. After experiencing it, a Xiaohongshu travel blogger believes that Xiaohongshu can meet basic search needs and the provided content can link to the original notes, helping users confirm the source of information.
(Part of the user's actual test post. Image source: Xiaohongshu screenshot)
However, there is room for improvement in continuing conversation ability, logical ability, and other aspects. At the same time, compared with similar products in China, "DAVINCI" still has some distance to go. For example, most similar products in China, such as Wenxin Yiyuan, can use plugins and have capabilities such as text-to-image and text-to-video, while DAVINCI's functionality is relatively single.
Xiaohongshu's layout of large models: The advantages are obvious, but the strategy is not clear
Since the beginning of the year, while major companies and entrepreneurs have been refining their models, Xiaohongshu has been taking a steady approach.
In March of this year, according to 36Kr, Xiaohongshu prepared an independent large model team, with core employees from the NLP technology team of the advertising business. The main person in charge is Zhang Debing, who has extensive experience in visual algorithm field and graduated from the Qiu Chengtong Mathematics and English Talent Class at Zhejiang University. He previously served as the head of the multimodal intelligent creation group at Kuaishou.
In April, Xiaohongshu launched an AI creation application called "Trik," which focuses on AI drawing. In July, Xiaohongshu launched graffiti and the "Now" function.
From Zhang Debing's recent sharing at the Xiaohongshu REDtech Youth Technology Salon, it is not difficult to see that Xiaohongshu's large model layout is mainly focused on three directions: First, the specific landing of multimodal technology in internal application scenarios at Xiaohongshu. Second, reducing costs and increasing efficiency for content creators through AIGC. Third, paying more attention to costs.
As a content community, Xiaohongshu's advantages in large model layout are obvious. First, Xiaohongshu has a large amount of high-quality graphic and textual data, note comments, and external open-source language materials, and the multimodal data foundation can help Xiaohongshu explore the possibilities across modalities. Second, Xiaohongshu also has certain algorithmic advantages in the search and recommendation scene.
By exploring the correlation between multimodalities and combining behavioral signals from user experience, Xiaohongshu can also drive continuous algorithm optimization.
It can be seen that Xiaohongshu's "not in a hurry" nature also reflects its uniqueness as a content community—decentralized distribution mechanism, UGC content ecology, and user experience are all core lifelines.
Xiaohongshu's Vice President of Technology, Wang Xiaobo, also stated that Xiaohongshu's layout of large models focuses more on products and applications and remains relatively calm. He believes that it is uncertain whether there is a universal base large model. Is it a breakthrough to build a large model of 100 trillion or 10 trillion? It is particularly uncertain now.
"Uncertainty" actually means that Xiaohongshu has not yet specifically figured out how to build large models. At present, its attempts are similar to what major companies are doing, and major companies are waging a "focus war" at the application layer.
Recently, ByteDance established a new AI department, Flow, with the technical leader being Hong Dingkun, Vice President of Technology at ByteDance, and the business leader being Zhu Wenjia, head of the large model team at ByteDance. Flow is an innovative AI business team under ByteDance and has currently launched two products, Dou Bao and Cici. Next door, Taotian is also establishing a large model research team, led by Taotian Group CEO Dai Shan, Taotian Group CTO Ruo Hai, and Alibaba Mama CTO Zheng Bo. Its landing areas are also Xiaohongshu's territory: search and recommendation, and contentization of browsing.
In addition to a more intense competitive environment, Xiaohongshu is also troubled by AI infringement and related issues. In August of this year, Xiaohongshu's AI drawing product TriK AI was accused of infringement by drawing creators. Illustrators stated that their original works were used for AI products without their permission.
(Creators sue Xiaohongshu's AI model library for infringement. Image source: Weibo screenshot)
The latest development in this incident is that four creators, @是雪鱼啊@画画的云淡风轻@正版青团子 @RedMatcha, have separately filed lawsuits against Xiaohongshu's main company, Xingyin Information Technology (Shanghai) Co., Ltd., and the main company of Xiaohongshu's Trik software, Ipsilong Information Technology (Beijing) Co., Ltd., and the case has been filed and is awaiting trial at the Beijing Internet Court.
In the past, major companies once besieged Xiaohongshu, solely because of Xiaohongshu's unique community genes. In the layout of large models, Xiaohongshu has also adopted a "defensive" strategy as it did in the past in commercialization.
However, this time, in an era where AI has not yet brought actual revenue to major companies, is Xiaohongshu's defense still a victory at a new starting point?
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
