Xiaohongshu Tests Self-developed AI Model “Little Sweet Potato” (Xiaodigua)
According to multiple independent sources obtained by Chinese media 36Kr, a large model team led by Zhang Debing, the head of AI innovation team at Chinese social media Xiaohongshu, has conducted gray-scale testing on their self-developed large model base “Little Sweet Potato”, or “Xiaodigua” in some internal products. The exploration of Xiaohongshu’s AI products is mainly overseen by Deng Chao, the person in charge of Xiaohongshu’s products and design.
In October 2023, at the REDtech Youth Technology Salon held by Xiaohongshu, Zhang Debing publicly mentioned for the first time two directions for the application of Xiaohongshu’s large model: multimodal technology and AI content creation tools. Over the past year, the layout of Xiaohongshu’s AI products has closely revolved around these two directions. For example, in July 2023, the AI doodle function and the AI illustration function “This Moment” were launched in the note publishing section.
On October 24, 2023, Xiaohongshu applied to register “Little Sweet Potato” as a trademark for its self-developed large model. Several informed sources told 36Kr that several AI products have already accessed the API of “Little Sweet Potato” for testing. According to one insider, the performance of “Little Sweet Potato” in application has reached the level of mainstream large models in China.
It is revealed by insiders that Xiaohongshu has been very cautious about launching new AI products because they fear disrupting the existing content ecosystem. One insider told 36Kr, “The content ecosystem of Xiaohongshu is about real people sharing their favorite finds. How AI can fit into this without causing discord is a problem that troubles Xiaohongshu.”
At present, the business focus of Xiaohongshu’s products is still on short videos, which occupy most of the human and financial resources, leaving not much resources for AI exploration. Several insiders told 36Kr that for Xiaohongshu, which just became profitable last year, its commercialization capabilities still need to be further proven. Before verifying the application scenarios, entering the expensive game of large models carries a high risk.
