TL;DR:
- SoftBank and KT, leading telecom companies, are making strides in the AI sector.
- SoftBank deploys an advanced IT platform with Nvidia AI supercomputing for GenAI development.
- SB Intuitions, a SoftBank subsidiary, focuses on creating large language models tailored for Japanese.
- KT introduces ‘Mi:dm,’ a versatile GenAI service for enterprises.
- KT’s AI stack eliminates the need for separate developments and offers data sovereignty.
- KT’s hyperscale AI computing service reduces GPU learning costs and enhances AI reliability.
- Both companies aim to lead in AI development, catering to diverse markets.
Main AI News:
In a bid to assert their dominance in the realm of artificial intelligence (AI), Japanese telecommunications powerhouse SoftBank and South Korean counterpart KT have made significant strides with their in-house generative AI (GenAI) systems. Both telecom giants aim to position themselves as leaders in the AI sector, showcasing their commitment to innovation and technological advancement.
SoftBank, for its part, has initiated the deployment of its cutting-edge IT platform, the linchpin of its latest GenAI endeavors. This formidable platform boasts a state-of-the-art Nvidia AI supercomputer, equipped with over 2,000 graphics processing units (GPUs) and powered by Nvidia’s enterprise-grade AI software.
SB Intuitions, a recently established subsidiary of SoftBank dedicated to research and development in large language models (LLMs), is harnessing this computing powerhouse. With a focus on creating homegrown LLMs tailored for the Japanese language, SB Intuitions is poised to embark on a comprehensive development journey. The ambitious goal is to construct LLMs with a staggering 350 billion parameters by 2024. Following this achievement, SoftBank plans to offer its formidable computing platform to universities, research institutions, businesses, and other entities at an early juncture.
With an unwavering commitment to high-speed processing, an abundance of engineering talent, and an extensive customer network, SoftBank is determined to craft homegrown LLMs enriched with Japanese language datasets. These LLMs will cater specifically to Japanese business practices and culture, creating a niche for generative AI services in the Japanese market.
Junichi Miyakawa, President and CEO of SoftBank, underscores the transformative potential of GenAI, stating, “GenAI is a groundbreaking technology that has the potential to fundamentally transform society.” He points out that the trend of foreign companies leading investments in computing platforms and LLMs has left Japanese AI development lagging. In response, SoftBank is pioneering its advanced, large-scale computing platform and homegrown LLMs, aligning them with Japanese culture and business practices.
While SoftBank leads the charge, South Korean telecom titan KT is not far behind. KT has introduced its GenAI service, ‘Mi:dm,’ which is designed to cater to enterprises of varying sizes and needs. This comprehensive offering encompasses a range of language models, from “lightweight” to “extra-large,” offering versatility to suit diverse company requirements.
KT’s AI stack eliminates the need for businesses to undertake separate developments and support learning infrastructure, making it a cost-effective choice. The telecom company emphasizes that its offering is domestically produced, reinforcing data sovereignty, and introduces ‘KT Mi:dm Studio,’ a customized environment for companies to access and utilize the ‘Mi:dm’ foundation model.
KT places a strong emphasis on security, addressing concerns that data may be controlled by tech giants. Furthermore, it claims that the majority of companies cannot afford to develop their own parameter models. KT’s foundation model, having completed pre-learning with trillion-level data, opens the door to accelerated business-to-business commercialization of LLMs, leading to a proliferation of AI business models and applications.
Notably, KT’s hyperscale AI computing (HAC) service reduces GPU learning costs by nearly 30% compared to existing products. It also boasts the ability to mitigate “hallucinatory” responses generated by AI, enhancing reliability in search, reasoning, and answering processes. These advancements are estimated to reduce the “hallucination phenomenon” by up to 70% compared to general GenAI services.
Initially targeting the business-to-business (B2B) market, KT plans to expand its AI offerings into five key sectors: global, manufacturing, finance, public, and education. Additionally, the company intends to foster innovation through an open startup ecosystem, poised to capitalize on the forecasted growth of the domestic AI market, projected to reach 4.4tn Korean won ($3.2bn) by 2027, according to IDC Korea.
These ambitious endeavors in the AI domain are not unique to South Korea alone, as telecom operators worldwide are gearing up to seize the burgeoning opportunities. A prime example is the collaboration between South Korean telco SK Telecom and German peer Deutsche Telekom, who plan to launch a multilingual language model by early 2024 as part of the Global Telco AI Alliance. This alliance, initiated in July, also includes Middle Eastern operator giant e& and Singaporean telco Singtel, signifying a global push towards AI innovation.
Conclusion:
SoftBank and KT’s significant investments in AI signify their commitment to becoming leaders in the telecom AI sector. SoftBank’s advanced IT platform and SB Intuitions subsidiary aim to revolutionize the Japanese AI landscape. KT’s ‘Mi:dm’ service, with its cost-effective and domestically produced AI stack, is set to empower businesses and accelerate AI adoption. As these telecom giants advance their AI capabilities, the market can expect increased competition and innovation, ultimately benefiting consumers and businesses alike.
