NVIDIA AI Days, hosted around the world, draw hundreds of enthusiasts, developers, researchers and startups to discuss and explore the latest technologies making AI breakthroughs possible.
NVIDIA AI Day Tokyo, held in late September, provided some insights on Japan's focus on AI. “Japan will see a 320x increase from 2020 in demand for AI computing power by 2030,” Kuniyoshi Suzuki, Senior Director of the cloud AI service division at SoftBank Corp., said at the event.
“To ensure transparency and safety as AI adoption expands, it is crucial to build a foundation of domestic technologies — high performance, Japan-made large language models (LLMs) and large-scale domestic computing infrastructure capable of continuous LLM development.”
“Specialised AI for industries like manufacturing, finance and healthcare will drive Japan’s digital transformation,” said Kazuya Ishikawa, evangelist of AI at NEC, a leader in IT and network technologies.
“We are certain that LLMs such as NEC cotomi enable professional employees’ knowledge transfer or utilising complex enterprise documents, directly addressing the skill gap and labour shortages that Japan is facing.”
Key announcements at the event included:
- Stockmark released a100-billion-parameter full-scratch Japanese LLM as an NVIDIA NIM microservice, offering 2.5x faster inference. Full-scratch refers to building an LLM from scratch, instead of using shortcuts such as existing models or high-level libraries.
- FastLabel launched FastLabel Data Curation, a solution for developing autonomous-driving capabilities and advanced driver-assistance systems.
- Hakuhodo Technologies, a service company of Japan’s major ad agency Hakuhodo, announced it will use NVIDIA AI Blueprints and the NVIDIA NeMo Agent toolkit to develop AI agents for autonomously producing advertisements.
- Shimizu Corporation, a general contractor with over 200 years of history in Japan, is exploring the NVIDIA AI Blueprint for video search and summarisation to monitor work progress and potential risks at construction sites.
 |
| Source: NVIDIA. Atsunori Fujita, Senior solution architect, NVIDIA Japan, introduced the new Nemotron-Personas-Japan dataset. |
The newly released Nemotron-Personas-Japan was also highlighted at AI Day Tokyo. The dataset is the first open synthetic dataset aligned with Japan’s real-world demographic, geographic and cultural distributions, providing a privacy-preserving, regulation-ready foundation for sovereign AI systems that does not rely on personal data.
“By leveraging advanced methodologies and accelerated computing, developers and businesses in Japan can empower AI agents to maximise enterprise data usage, drive organisational efficiency and achieve continuous innovation,” said Bartley Richardson, Senior director of engineering at NVIDIA.
“Japanese developers are building flexible and powerful agentic systems suitable for Japan’s drive toward technological excellence and operational precision.”
In 2024, Japan announced plans to invest at least US$65 B through fiscal year 2030 to boost the semiconductor and AI industries. This year, the innovation-first Act on Promotion of Research and Development, and Utilization of AI-related Technology (人工知能関連技術の研究開発及び活用の推進に関する法律案要綱) or Japan AI Act was enacted. Japan’s GENIAC initiative also aims to strengthen domestic generative AI capabilities by providing companies with computing resources, fostering collaboration and supporting foundation model development, including LLMs tuned for the Japanese language and industry.
No comments:
Post a Comment