Konan Technology, LLM Technology and Business Experience at the Highest Level.. Confident as a National AI Representative [Meeting the AI Innovator]

"A model that has both continuous performance enhancement and continuous product release, along with extensive business performance, is unique to Konan Large Language Model (LLM). If we can additionally secure 1000 graphic processing units (GPUs), we are confident that we can create a national-level AI model based on accumulated development know-how."
Kim Young-seop, CEO of Konan Technology (photo), said this in an interview with Financial News on the 28th after declaring his candidacy for the 'Independent Artificial Intelligence (AI) Foundation Model Development Project' hosted by the Ministry of Science and ICT. 
The Ministry of Science and ICT plans to make the AI foundation model developed by AI developers through this project available as open source if it reaches a global level. Resources needed for AI model development, such as graphic processing units (GPUs) and datasets, will be supported according to demand. If selected as a representative AI model, names such as 'K-AI Model' and 'K-AI Company' can be used. The Ministry of Science and ICT reduced the number of main bidders from 15 to 10 through a document review last week. At the end of August, 5 elite teams will be selected. Currently, the 10 selected include domestic big tech companies, telecommunications companies, startups, and academia, with Konan Technology being one of the 10 survivors. 
Konan Technology is the main company that formed a consortium with three AI companies, Synapsoft, Alchera, and Pebblus, and four universities, Korea University, Yonsei University, POSTECH, and KAIST.  The participating faculty has published more than 180 SCI papers in the AI field and more than 210 papers at top AI conferences in the past five years, indicating a very solid theoretical foundation in the foundation model field, according to the company. In particular, Konan Technology emphasized its accumulated technology, business performance, and personnel expertise as strengths. 
CEO Kim said, “Konan Technology started developing the foundation model itself in December 2022, and introduced NVIDIA H100 graphic processing units (GPUs) for the first time in Korea to speed up development, showcasing Konan LLM with 13.1 billion and 41 billion parameters in 2023.” He explained, “In July last year, it was listed on the Stanford University Ecosystem Graph, recognizing its global competitiveness, and in March this year, it launched the ENT-11 model, the first in Korea to integrate inference functions.”
He added, “Together with Korea Southern Power, we completed the development of the first universal generative AI chatbot for public institutions, ‘KEMI,’ this month, and recently won major LLM projects for institutions such as the ‘Gyeonggi-do Specialized LLM Construction’ for Gyeonggi Province and the ‘Legal Specialized LLM Development’ for the Supreme Court.” He stated, “We have experience in various construction and business proof-of-concept (PoC) projects, including more than 20 cases in the private sector, in addition to major public institutions.” Recently, the company also integrated the ‘Konan AI Station Server’ equipped with LLM into the neural processing unit (NPU) of AI semiconductor company Rebellion.
Konan Technology Consortium plans to spread the AI transformation (AX) platform for manufacturing to industrial complexes and enhance AI accessibility to elementary, middle, and high schools nationwide through the school AI platform, while promoting the overseas expansion of sovereign AI through cooperation with Southeast Asian countries. Through this, they aim to strengthen the competitiveness of the domestic AI industry and create an ecosystem that can lead to global market entry. 
CEO Kim emphasized that Konan Technology has experience with an H100 cluster of about 140 units and the learning and processing of high-quality Korean data owned by the company, amounting to 25 billion cases, and that if selected as a national AI company, it would be an immediate power source capable of rapid response. 
He mentioned, “With 140 H100 units, we developed the 13.1 billion, 41 billion, and 3 billion parameter LLMs for on-device use, as well as the first inference model in Korea, the ENT11 model, in just two months,” adding, “The inference model was released faster than Naver.” He continued, “Konan Technology has been developing domestic AI technology for 26 years with a mission to increase the nation’s AI independence, and we have prepared sincerely for this project.” He pledged, “We are very grateful for the Ministry of Science and ICT’s promotion of this project, and if selected, we will contribute to advancing Korea as one of the top three AI powerhouses.”
solidkjy@fnnews.com Koo Ja-yoon Reporter