The Tokyo Institute of Technology and Fujitsu announced their collaboration on May 22 to develop foundational technology for generative artificial intelligence (AI) using Japan's supercomputer Fugaku.
The project aims to foster the development of domestic generative AI, enhance AI-related technologies in Japan, and strengthen the country's global competitiveness. By the end of FY 2023, the team intends to establish a technology that primarily learns from Japanese-language text data. They further aim to make it available to domestic companies and universities by FY 2024, free of charge.
OpenAI's ChatGPT and similar generative AI tools can produce diverse content through extensive training on data. However, their heavy dependence on English text data results in incorrect or inaccurate responses to Japanese inquiries and topics.
Japan's Own Chatbot
To overcome this limitation, the team at Tokyo Tech is focused on developing a "large-scale language model" as the foundation for generative AI. This utilizes Japanese-language text data as the primary source.
They plan to train the AI tech using Japanese news articles, blog posts, and online encyclopedia data, including sources like Wikipedia. Collaborations with RIKEN, the operator of Fugaku, and Tohoku University are integral to the project. In addition, they are also exploring potential partnerships with Nagoya University and CyberAgent, a major media company.
One of the team members is Associate Professor Keisuke Sakaguchi, who specializes in information sciences at Tohoku University. He says, "Relying heavily on foreign technologies like ChatGPT developed in the United States would leave Japan unable to innovate independently. Therefore, Japan must gather expertise and knowledge [in AI development] to acquire technologies for domestic production."
You must be logged in to post a comment Login