Japan’s NEC on generative AI’s vanguard

The hard work of building secure and reliable systems for numerous practical applications is where the future and promise of generative artificial intelligence (AI) lies. Japan gets it; many in the West do not.

Japan’s NEC has spent nearly three years building its own AI supercomputer, constructing the best-performing Japanese language large language model (LLM) while testing other AI products in its offices and factories. Once they work properly, they will be marketed to paying clients.

NEC is Japan’s leading supplier of computing and communications equipment and a leading provider of software-based services for business, industry and social infrastructure.

An LLM is an AI algorithm that uses deep learning (machine learning based on neural networks) and an extensive database to assemble information and generate human language-based text at great speed.

NEC executive vice president and chief technology officer (CTO) Motoo Nishihara recently told the Nikkei newspaper that his company has developed the AI capacity to cut the time required to prepare documents in half and speed up the transcription of meeting notes from 30 minutes to five since trials were begun in May 2023.

NEC also reports reductions of up to 80% in the number of work-hours required to create source code for internal system development. The company’s generative AI system is being used about 10,000 times a day by approximately 20,000 employees, according to reports.

NEC’s AI supercomputer became operational in March 2023. Equipped with 928 graphics processing units (GPUs) from Nvidia plus central processing units (CPUs) from Intel, it supports NEC’s proprietary LLMs and other AI operations with reputedly the greatest computing power among Japanese