Group starts developing generative AI well-versed in Japanese

A Japanese group of universities and entities announced on Monday that they have started developing "Large Language Models," or LLMs, to be used in generative AI systems. They hope the work will lead to AI that is specifically good at handling the Japanese language.

LLMs serve as the basis of ChatGPT and other AI that can automatically generate sentences and images.

Researchers from the Tokyo Institute of Technology, Tohoku University, electronics firm Fujitsu and the Riken research institute plan to create LLMs by the end of March next year, using Japan's supercomputer Fugaku.

In other countries, engineers are developing LLMs that learn mainly in English. They are said to be bad at Japanese.

The Japanese group says Fugaku will facilitate the learning of a vast amount of Japanese sentences. It hopes the result will lead to the development of a generative AI well-versed in the Japanese language.

Last week, Japan's IT firm CyberAgent made public an LLM that had learned mainly Japanese. Messaging app operator Line has also been studying such a model.