Technology

Supercomputer Fugaku: Powering the Development of Japanese Generative AI

The Tokyo Institute of Technology, in collaboration with Riken research institute, Fujitsu Ltd., and Tohoku University, has embarked on a groundbreaking project to develop highly performing generative artificial intelligence (AI) using the powerful Fugaku supercomputer. This initiative aims to leverage Fugaku’s exceptional calculation speed to accelerate the development of generative AI and establish Japan’s position in the field of AI research. By inputting information in Japanese, the project seeks to address the limitations faced by existing generative AI tools, which often struggle to provide accurate information about Japanese history or culture.

Enhancing Generative AI with Fugaku

Generative AI, popularized by the U.S. startup OpenAI’s ChatGPT, has gained widespread attention for its ability to learn vast amounts of text data. However, most major generative AI tools primarily rely on English language data, leading to inaccuracies when queried about Japanese topics. The project aims to bridge this gap by leveraging Fugaku’s computational power to develop methods that enhance the performance of generative AI using Japanese language sources, including the Japanese version of Wikipedia.

Promoting Collaboration and Knowledge Sharing

The Tokyo Institute of Technology, Riken, Fujitsu, and Tohoku University have joined forces to develop generative AI capable of producing high-quality and accurate responses in Japanese. By openly sharing their development processes and outcomes, the researchers aim to facilitate the growth of homegrown AI in Japan. The project intends to publish the generative AI model and development methodology by the end of the fiscal year, enabling Japanese companies and research institutions to benefit from the findings.

Competing with Global AI Organizations

Beyond the immediate goals of improving generative AI accuracy and performance, the project also aspires to create AI systems capable of rivaling those produced by leading global organizations like Google. By nurturing AI research and development within Japan, the project seeks to foster innovation and address challenges encountered by existing generative AI models, such as occasional incorrect responses. The ultimate aim is to establish Japan as a formidable contender in the AI landscape.

The Power of Supercomputer Fugaku

Supercomputer Fugaku, housed at the Riken Center for Computational Science, has proven to be a formidable machine. With an impressive 7,630,848 cores, it held the top position in the global supercomputer rankings from June 2020 to November 2021. While it has since been surpassed by Frontier, Fugaku remains a significant asset in advancing scientific research and computational capabilities in Japan.

The collaboration between the Tokyo Institute of Technology, Riken, Fujitsu, and Tohoku University, supported by the powerful Fugaku supercomputer, heralds a new chapter in Japanese generative AI development. By focusing on Japanese language sources and addressing the limitations of existing generative AI models, the project aims to provide accurate and reliable AI systems. The open sharing of knowledge and methodologies will contribute to the growth of AI research in Japan, paving the way for the country to compete globally in the field of artificial intelligence. With Fugaku’s computational prowess and the collective expertise of the participating organizations, the project holds the potential to advance the frontiers of generative AI and propel Japan to the forefront of AI innovation.

 

Tags: Artificial Intelligence (AI)

One thought on “Supercomputer Fugaku: Powering the Development of Japanese Generative AI

Comments are closed.