Supercomputer Fugaku trains AI with Japanese model

Starting this month, the supercomputer Fugaku will be used to develop basic technology for generative AI with high Japanese language proficiency, a team from a Japanese research institute and Fujitsu Ltd. said on Monday.

In parallel with this development, the team will work on solving problems related to artificial intelligence models, such as piracy.

Tokyo Institute of Technology, Tohoku University, and government-funded RIKEN, in collaboration with the RIKEN Center for Computational Science and Fujitsu, which developed Fugaku, will set out to create a so-called large-scale language model (LLM).

Generative AI models such as OpenAI’s ChatGPT are currently learning mainly from online English teaching materials. Therefore, the Japanese produced by these models sounds unnatural or inaccurate.

AI learns a large amount of Japanese document data to improve Japanese language proficiency.

Since the development of AI requires a huge amount of calculations, the team aims to proceed efficiently by utilizing the world’s most advanced supercomputer “Fugaku” in Kobe.

LLM will be released free of charge by the end of March 2024 so that universities and companies can use it for basic research and new service development.

In the future, it is expected to contribute to the realization of domestic generation AI.

At the same time, we aim to establish technology to deal with personal information infringement and copyright infringement, which are concerns about generative AI.

Professor Rio Yokota of the Tokyo Institute of Technology, who is a member of the team, says, “We need to solve various problems in order for generative AI to become widespread in society.” “We will proceed while ensuring the transparency of the development process.”