ByteDance¶
On February 28, 2024, Interface News learned from multiple informed sources that ByteDance is secretly developing several products in the field of AI LLMs, including multimodal digital human products and AI-generated images and videos.
According to one insider, they saw a demo of ByteDance's multimodal digital human product in the second half of last year and felt it was quite good overall.
Additionally, Interface News has learned that ByteDance's subsidiary, CapCut, formed a closed team months ago to secretly develop AI products. Currently, this team is still in a strict confidentiality phase, and the products under development have not yet been launched.
Interface News reached out to ByteDance for confirmation of the above news, but as of publication, no response has been received.
An insider close to ByteDance stated that throughout last year, founder Zhang Yiming focused most of his energy on AI, indicating the company's high regard for its AI business.
Currently, ByteDance is taking a comprehensive approach to the research and development of AI LLM-related products, employing a multi-faceted strategy that spans from the model layer to the application layer.
In the foundational LLM field, last August, the company launched its first large language model "Doubao" and the multimodal LLM BuboGPT. Its Douyin Lark LLM has been registered under the first batch of "Interim Measures for the Management of Generative Artificial Intelligence Services" and is open to the public.
A few days ago, ByteDance also released the image generation model SDXL-Lightning, which can generate extremely high-quality and high-resolution images in 2 to 4 steps, accelerating the generation speed by ten times.
At the AI application layer, ByteDance established a new AI department called Flow in November last year, which has already launched three AI chat products: Doubao, Kouzi, and Cici. At the foundational LLM layer, ByteDance has made arrangements in both language and image modalities, with both teams reporting to Zhu Wenjia, the technical head of TikTok.
Another person close to ByteDance revealed that the company is currently facing considerable pressure in its LLM layout due to strategic oscillation between self-research and investment over the past year.
The insider stated that ByteDance initially planned to enter the LLM field through investments and once considered investing in the LLM companies MiniMax and LeapStar, but decided to abandon external investments in LLM companies in June of last year and shifted towards self-research.
"In self-research, ByteDance's progress has not been faster than that of startups. In terms of investment, especially after Alibaba's recent significant investment in the dark side of the moon, ByteDance's decision to completely abandon investment needs to be reassessed," the insider said.
However, several individuals familiar with ByteDance's LLM situation emphasized that it cannot be completely ruled out that the company has a layout in the AI LLM field. Among all ByteDance products, the most promising candidate to implement ByteDance's AI LLM is CapCut.
One insider analyzed that CapCut is a video creation tool situated upstream in content creation, and moving towards AI means generating videos from text. Additionally, the video content created with CapCut has a platform in Douyin, and creators using ByteDance's text-to-video and multimodal digital human products for content creation have significant potential.
Before this year's Spring Festival, former Douyin Group CEO Zhang Nan resigned from the CEO position, stating that he would focus on the development of CapCut in the future. This move has been interpreted by many industry insiders as ByteDance's intention to push forward in the text-to-video direction through CapCut.
"CapCut needs to first address the issue of creative materials, including various personalized materials related to video and animation," said the insider.