A new dawn for generative AI shaping the business landscape

NEC’s AI models are driving global economic growth and addressing AI deployment challenges. Its Generative AI promises a significant productivity revolution, thanks to its Japanese-proficient LLM. Plus, its AI solutions are reshaping businesses and amplifying global economic growth.

The groundbreaking technology of Generative AI is transforming not just routine tasks like data classification, but also creating unique experiences with its capability to generate human-like text, music, and digital art. This innovation is piquing consumer interest, leading to a diverse range of stakeholders exploring the impact of this AI form on business and society, often with limited foundational understanding.

The swift progression of generative AI intensifies this challenge. A testament to its rapid evolution was the release of ChatGPT in November 2022, which was quickly succeeded by a more advanced language model, GPT-4, from OpenAI just four months later. The technological marathon continued with Google unveiling several AI-driven features in May 2023, such as Search Generative Experience and a new Language Model, PaLM 2.

The potential of these breakthroughs to drive worldwide economic expansion is what truly distinguishes them. According to recent research by McKinsey, generative AI may generate an economic value from US$2.6 trillion up to US$4.4 trillion per year, spanning 63 different use cases. This underscores the monumental potential of such technology to act as a key instigator of global productivity.

NEC’s customized AI solutions for the future

Adding to this momentum is the proactive stance of NEC Corporation, a Japanese multinational IT and electronics company. NEC is redefining customer experience by creating customized generative AI solutions that can adapt to technological shifts and unlock novel business value. Starting this month, NEC’s Generative AI Service will begin offering proprietary language model licenses, dedicated hardware and software, consulting services, and more in Japan, with plans to extend to new markets beyond FY2024 gradually.

In a bid to foster collaboration, NEC is launching the “NEC Generative AI Advanced Customer Program” this July. This initiative targets collaborating with approximately 10 companies and universities to offer comprehensive support for language model utilization, from model development to the software and organizational framework necessary for efficient use.

Reinforcing its commitment to business transformation, NEC established the “NEC Generative AI Hub” on July 1. This specialized organization of researchers, consultants, digital trust specialists, and other professionals aims to generate sales of around US$346 million over the next three years from these AI-related ventures.

Despite the optimistic outlook, the application of generative AI presents its own challenges, particularly regarding security, reliability, and applicability. Beyond translation and summarization tasks, businesses must leverage AI to harness and apply unique company-specific knowledge internally.

Addressing these challenges, NEC has made substantial progress. In May, the corporation began using generative AI for internal purposes, successfully setting up a secure and user-friendly system within two weeks. This system is now being utilized by approximately 20,000 users, resulting in significant efficiency improvements, including reductions in source code creation time, document generation time, and meeting minutes recording time.

Building an AI business ecosystem

NEC refines the system further to optimize tasks common to all employees, such as document proofreading, and specialized tasks like software development. The insights gained from its internal use of language model-related services will be instrumental in achieving these goals.

NEC’s unique language model is notable for its excellent command of Japanese and compact structure, enabling it to operate on a standard GPU server. The “NEC Generative AI Service Menu” offers many customer support services, contributing to swift and effective business transformations. NEC also offers the “NEC Generative AI Framework” software to streamline the language model integration into customer workflows.

This AI service and the “Microsoft Azure OpenAI Service” can be customized to fit customer needs, including secure on-premises environments. Hosted primarily at the NEC Inzai Data Center, these services ensure a secure and efficient operation of the language models.

NEC’s lightweight model, boasting 13 billion parameters, is designed to function smoothly in cloud and on-premises environments. It excels in Japanese language comprehension, as demonstrated by a business-related knowledge and reading comprehension benchmark test.

The model is already enhancing internal operations significantly, improving efficiency in tasks such as document generation and coding assistance. While most language models are primarily focused on English, NEC’s model stands out as a high-performing Japanese model, adaptable across industries and poised to accelerate business applications and boost productivity.

Some of the features include:

• Japanese language proficiency: NEC’s LLM demonstrated superior performance on the JGLUE benchmark, achieving an 81.1% accuracy in question-answering and surpassing leading competitors with 84.3% in reading comprehension, indicating its potential to excel across industries.
• Compact structure: Despite its high performance, NEC’s LLM requires only a single GPU, making it efficient and cost-effective. It can also be swiftly built and deployed in on-premises environments, ensuring security for sensitive operations.

NEC achieved this superior performance by utilizing high-quality data and computation time, while keeping the parameter size manageable. The company constructed this high-performance language model using its largest AI research supercomputer in about a month.

Looking ahead, NEC plans to use the base model to create company-specific language models using private data, committed to improving performance and hastening practical applications through the NEC Generative AI Hub.

---

---

---

---