Generative AI Trials to Streamline Development and Maintenance Operations by Fujitsu and Mizuho

Fujitsu and Mizuho Financial Group, Inc. (hereinafter Mizuho) launched joint verification trials using Fujitsu's generative AI technology to streamline development and maintenance operations of Mizuho's system. The trials will take place between June 19, 2023 and March 31, 2024.
Generative AI technologies offer enormous potential to transform the ways we work, and with rigorous verification may soon allow companies to optimize various work processes and tasks.
Recognizing the promise of this technology, Mizuho is now exploring ways to incorporate AI in its operations to improve the quality and resilience of its systems during the development and maintenance phase.

Fujitsu is actively promoting research and development into generative AI models to support system development and maintenance technology under the framework of its "Fujitsu Kozuchi (code name) - Fujitsu AI Platform," which supports users globally to accelerate the testing and deployment of advanced AI technologies.

Within their joint trials, Fujitsu and Mizuho will focus on applying Fujitsu's generative AI technology within the development process of Mizuho's system, with the aim of verifying the AI's effectiveness in automatically detecting errors and omissions in system design plans and audit processes.

Fujitsu started offering an AI core engine for generative AI on June 9, 2023 via "Fujitsu Kozuchi (code name) - Fujitsu AI Platform." The generative AI technology used in the joint trials combines this AI core engine for generative AI with Fujitsu's expertise in engineering.
The two parties further plan to develop a technique to automatically generate test specifications from system design plans improved with the aforementioned technology. To strengthen work process resilience, Fujitsu and Mizuho further plan to apply generative AI technologies to work processes and tasks including the generation of source code, system development, and maintenance.

Fujitsu, in cooperation with Tokyo Institute of Technology (Tokyo Tech) (1), Tohoku University (2), and RIKEN (3) in May further announced the start of a R&D project of a distributed training of Large Language Models (LLM) (4) on supercomputer Fugaku, with plans for release in fiscal 2024.

Fujitsu will leverage research results from this project to promote practical research in the industrial world and apply them to Mizuho's system to further improve quality and efficiency in Mizuho's operations.

(1) Tokyo Institute of Technology:
Meguro-ku, Tokyo, Japan; President: Kazuya Masu
(2) Tohoku University:
Aoba-ku, Sendai, Japan; President: Hideo Ohno
(3) RIKEN:
Headquarters: Wako City, Saitama Prefecture, Japan; President: Makoto Gonokami
(4) Large Language Models (LLM) :
Neural networks with hundreds of millions to billions of parameters that have been pre-learned using large amounts of data. Recently, GPT in language processing and ViT in image processing are known as representative large-scale learning models.