KT, Microsoft to develop a Korean ChatGPT by 2025
Published: 10 Oct. 2024, 18:06
- LEE JAE-LIM
- lee.jaelim@joongang.co.kr
KT, Korea’s major telecommunications provider, and Microsoft aim to complete the development of a Korea-customized version of ChatGPT by the second quarter of 2025 under a partnership that will oversee the investment of 2.4 trillion won ($1.8 billion) over the next five years.
The two firms first secured a strategic partnership in June related to AI and cloud computing, then later announced in September that they would co-develop a customized version of GPT-4o, a faster and more capable model of OpenAI's flagship GPT-4 model. At a press event in central Seoul on Thursday, the companies clarified that they plan to complete the tailored AI model by the second quarter of next year.
KT also plans to release a small language model (SLM), based on Microsoft's Phi-3.5, which is tailored to the public and financial sectors, on the same timeline. The mobile carrier is currently in the process of training the chatbot on data it has obtained in the realms of education, history and culture.
The company expects the partnership to generate 4.6 trillion won of revenue, encompassing 269 billion won in 2025, 610 billion won in 2026, 1.1 trillion won in 2027, 1.3 trillion won in 2028 and 1.37 trillion won in 2029.
Half the $1.8 billion investment will be allocated to data center infrastructures and graphic processing units (GPUs), and the remaining 50 percent will go toward research and development (R&D) as well as marketing and developing the Korea-customized AI models.
“Microsoft understands the operational mechanisms of enterprises better than any other company,” KT CEO Kim Young-shub said at Thursday's event. “Our discussion with Microsoft, which began in December of last year, has only strengthened this belief. Microsoft’s capabilities in AI development are unrivaled.”
The two companies will jointly develop and release a cloud service for public and financial sectors in the first quarter of next year, taking into account Korea’s regulatory and security environment.
However, SLMs for enterprises will be backed by KT’s proprietary large language model Mi:dm.
KT and Microsoft will also establish a joint venture as a KT subsidiary for AI transformation in the first quarter of next year, consisting of employees in the AI and cloud sectors. They will also co-establish a research center to advance the technologies next year.
KT rebutted speculation that the subsidiary will merely function as a company that sells Microsoft products, stating that the purpose of the venture is to attract top talent to resolve clients’ problems.
“The goal is to build a strong foundation, attracting top global experts to boost our AI ecosystem,” Kim said. “Overseas expansion is a long-term goal, but we will be focusing on strengthening our capabilities first.”
In the cloud sector, KT aims to become a company that can juggle roles of a cloud service provider (CSP) and managed service provider (MSP). Its cloud subsidiary, KT Cloud, will resume its position as a CSP and collaborate with Microsoft in the domain of secure public cloud.
KT will also secure Nvidia’s AI chips during the partnership period.
“We have secured access to GPUs and neural processing units (NPU) in Korea with Microsoft,” said Jung Woo-jin, executive vice president of KT's consulting unit's tech innovation group. “You may have seen the recent news that the H200, Nvidia’s latest GPU, will first be supplied to Microsoft. We are working with Microsoft to be the first to receive supply in Korea. Moreover, discussions are also underway for enhanced partnerships based on the GPU's demand and supply.”
BY LEE JAE-LIM [lee.jaelim@joongang.co.kr]
with the Korea JoongAng Daily
To write comments, please log in to one of the accounts.
Standards Board Policy (0/250자)