TL;DR:
- Chinese tech unicorn 01.AI decided to alter the tensor name of its AI model, Yi-34B, due to its reliance on Meta Platforms’ Llama system.
- 01.AI describes the renaming as an “oversight” and clarifies that it was not an attempt to mask the source of the AI model.
- The incident reflects the complexity of AI model development in China and the push for national AI standards.
- Baidu CEO criticizes the rush to develop AI models, while the open-source community sees no issue in sharing code and ideas.
- The retention of tensor names linked to the Llama architecture is seen as valuable in the open-source community.
- 01.AI, backed by Alibaba and Sinovation Ventures, is now among China’s leading AI unicorns.
Main AI News:
In a recent development, Chinese tech unicorn 01.AI, founded by Taiwanese venture capitalist and computer scientist Lee Kai-fu, has acknowledged an oversight in its decision to change the name of its open-source artificial intelligence (AI) large language models (LLM) Yi-34B. This decision came after an inquiry revealed that the model was based on the architecture of Meta Platforms’ Llama system.
The Beijing-based 01.AI, which has seen its value soar to over US$1 billion in less than eight months since its inception, has characterized this matter as “an oversight.” In response, the tensor name of its LLM, the technology utilized for training intelligent chatbots such as ChatGPT, will be modified to accurately reflect its reliance on Meta’s LLM. This announcement was made via a post on the Hugging Face open-source community platform by 01.AI’s open-source director, Richard Lin, in response to an inquiry by AI researcher Eric Hartford.
Tensors, crucial data containers in the AI machine-learning process, organize and store information systematically, facilitating LLMs in comprehending and generating human-like text. LLMs, on the other hand, are deep-learning AI algorithms capable of tasks such as recognition, summarization, translation, prediction, and content generation using extensive datasets.
According to Lin, “During extensive training experiments, we made several renamings in the code to meet experimental requirements. But we kinda dropped the ball and didn’t switch them back before pushing out our release … We’re sorry for the confusion.” 01.AI clarified that the renaming of the tensor was done to thoroughly test the Llama model, with no intention of concealing the AI model’s source.
This oversight by 01.AI sheds light on the intricate landscape of activities driving the rapid development of various LLMs in mainland China. The government has established a new regulatory body in Beijing as part of efforts to establish a national standard for AI models.
This incident coincides with Baidu CEO Robin Li Yanhong’s criticism of the frenetic development of LLMs by Chinese tech companies as “a huge waste of resources.” In contrast, Hugging Face community member Hartford, who raised questions about the tensor name of 01.AI’s Yi-34B LLM, emphasizes the norm of sharing code and ideas within the open-source community. He points out that while various AI models may share the same architecture, they are trained with distinct data sets.
Moreover, Hartford suggests that retaining the same tensor names as the Llama architecture holds value due to investments and tooling associated with it. This approach aligns with honoring conventions within the open-source community, similar to citing research papers and providing proper attribution.
01.AI, backed by Alibaba Group Holding’s cloud unit and Sinovation Ventures, has emerged as one of China’s leading AI unicorns, alongside companies like Baichuan, founded by ex-Sogou founder Wang Xiaochuan, and state-backed ZhipuAI.
Conclusion:
This incident highlights the challenges and nuances in the fast-paced world of AI model development, particularly in China. While 01.AI’s renaming of its AI model may have been an oversight, it underscores the need for transparency and adherence to open-source community conventions. It also reflects the broader context of China’s drive to establish national standards for AI models. As the AI market continues to evolve, companies should prioritize ethical practices and collaboration within the open-source community to maintain trust and credibility.