Oct 31, 2023

The Korea Times Posted : 2023-10-31 16:09 Updated : 2023-10-31 16:39 Choi Joon-ki, second from left, senior vice president and head of KT'’s AI and big data division, speaks during an online press conference introducing the company’s hyperscale AI platform Mi:dm, Tuesday. Courtesy of KT Telecom company aims to expand into B2B AI market By Baek Byung-yeul KT unveiled its hyperscale AI platform Mi:dm and will expand its AI business in business-to-business (B2B) markets at a time when increasing numbers of organizations are considering adopting the emerging technologies to improve efficiency and productivity, the telecom company said Tuesday. The company said its AI platform consists of four variations, such as basic, standard, premium and expert, and that enterprise customers can choose the appropriate level to match the size of their organization and purpose of use. With the hyperscale AI platform, KT will try to expand into the global AI market, manufacturing, finance, public agencies and education. “Our business strategy for Mi:dm is to focus on the B2B market and lead the era of large AI models becoming mainstream for enterprise customers,” Choi Joon-ki, senior vice president and head of KT's AI and big data division, said during an online press conference. "We want to provide hyperscale AI models that are tailored to enterprise customers, and we want to provide models with expertise in specific markets.” Choi added that KT will open the foundation model of Mi:dm to all companies that want to use and learn hyperscale AI. The foundation model is the core underlying model of a large AI model that has been trained on a massive dataset. Using this foundation model, companies can fine-tune and develop various AI application services. "We will open the foundation model of Mi:dm AI to create an AI ecosystem," he said. "Also, we will continue to explore monetization models that can be used with large AI, such as global expansion." To make it easy for enterprise customers to develop AI services, KT will open a service called Mi:dm Studio to provide customers with a convenient development tool. Another advantage of KT Mi:dm is that it provides a reliable package designed to reduce false responses known as hallucinations. Hallucination refers to AI's creation of plausible content, presented as truth, that is incorrect or nonsensical. KT developed technology to improve reliability at all stages, including search, inference and response. KT explained that it can reduce the hallucination phenomenon of Mi:dm by about 70 percent compared to existing models through the technology. "We provide Mi:dm Studio so that any customer can easily use large AI. In addition, we need to solve hallucinations, so we have strengthened the reliability of the answers of the large language model (LLM) by providing a reliable package that we have developed internally," the head of the AI and big data division said. To expand the AI markeplace in this industry, KT will jointly develop AI business models using Mi:dm with various AI startups, such as Upstage for enterprise-only LLM business, Qanda and Enuma in the education field and BI Matrix in the personal assistant field for B2B market, to target domestic and global markets. "We partnered with KT because the company has the best cloud computing business and the best capabilities in B2B services and it has hyperscale AI Mi:dm," Kim Sung-hoon, CEO of Upstage, said. "We will work with KT and KT's partners to create overwhelming AI models, complete a variety of AI portfolios, and become the global leader."