Journalists from the Dong-A Ilbo IT Science Team introduce noteworthy technologies, trends, and companies in the fields of IT, science, space, and bio. “What is this company?” The behind-the-scenes stories of tech companies changing the world with technology! From ideas that surprised the world to the current concerns of founders, they delve into everything about them.
The decisive battle to determine the 'National Representative Artificial Intelligence (AI)' has begun in earnest. Five elite teams (NC AI, Naver Cloud, SK Telecom, LG AI Research Institute, Upstage) supported by the government to develop independent AI foundation models must aim to develop AI models and services that perform at least 95% of the latest global AI models over the next two years. By the end of 2026, the top two passing public evaluations will be selected.
Among the 'Top 5' selected by the government on the 4th, NC AI received the highest interest from the industry. This is because NC AI, a subsidiary of NCSoft known for the online game 'Lineage', made it to the top 5, surpassing prominent tech companies and telecoms like Kakao and KT. However, NC AI does not view this selection result as an 'upset'. It is confident that it is the result of 14 years of steadfast investment and research in AI.
Kim Geon-soo, head of the Agentic AI Lab at NC AI, who led this project, emphasized in an interview with the Dong-A Ilbo at the NC Research and Development (R&D) Center in Seongnam, Gyeonggi Province, on the 7th, “There are almost no companies in Korea other than us that have experienced the full stack (entire process) from AI model research to service development and operation.” NC AI began by releasing Barco LLM (Large Language Model) in August 2023 and is developing and unveiling the Barco series. Last month, it launched 'Barco Vision 2.0', a multimodal AI that simultaneously understands images and text, continuing the innovation of providing general-purpose AI to the industry as a game company for the first time in Korea.
Kim Geon-soo, head of the Agentic AI Lab at NC AI, is having an interview with the Dong-A Ilbo at the NC Research and Development (R&D) Center in Seongnam, Gyeonggi Province, on the 7th. Provided by NC AI
The consortium led by NC AI made it to the top 5. While Kakao and KT were eliminated, significant achievements were made as a 'game company'. What was the differentiating point?
“The industry was surprised, saying, ‘Were they this prepared?’ We formed a ‘grand consortium’ that gathered the capabilities of the entire domestic and international AI ecosystem. Verified leaders from 54 industry-academia-research institutions joined forces. This allowed us to flawlessly implement the entire cycle from basic research to technology application, AI service, and global expansion.
It’s not just about having a large number of participants. When we announced our thoroughly secure strategy, the overwhelming scale surprised even those present. We partnered with leading companies from various industries, and SI companies like POSCO DX, which are engaged in AX (AI transformation) business, joined us, focusing on scalability.”
A ‘grand consortium’ was formed, involving 54 companies and institutions from academia, such as Seoul National University, KAIST, and the Electronics and Telecommunications Research Institute (ETRI), and the industry, such as Lotte Innovate and POSCO DX. What was the background for forming the ‘grand consortium’?
“It’s the trust in NC’s technology. NC AI has consistently accumulated technology in the fields of games, content, and industrial AI over the past 14 years. It is important that collaborative partners who know and trust our technological capabilities and AI development history participated.”
What was the core message of the ‘Sovereign AI’ strategy?
“We deeply pondered what Sovereign AI is. We concluded that Sovereign AI is the expansion of Korea’s industrial competitiveness.
To protect and develop the nation through Sovereign AI, we believe that achieving AX in industries we are already good at and incorporating our foundation model will help Korea become one of the top three AI powers.
While writing the business proposal, we were surprised to find that although many companies want to transition to AI, the actual transition rate was about 1%. This is directly related to our country’s productivity.”
Did you consider scalability for global expansion?
“We are also partnering with other leading domestic SI companies whose names cannot yet be disclosed. We see industrial scalability that can contribute to the globalization of domestic industries as a core focus. We believe that contributing to the AX of our companies and helping them enter the global market is a way to enhance national competitiveness.
To this end, we will build a ‘DomainOps platform’ to make AI models specialized for each industry easily usable. This means creating an ‘industry-specialized AI platform’ that can be widely utilized in industries where Korea can be the best globally. Overseas, there is also a need for manufacturing AX, and exporting our foundation model can establish a bridgehead for global expansion.”
“Goliath and David?
There was an evaluation that ‘NC AI’s capabilities are quite substantial’ upon closer inspection. Was there a secret weapon?
“It’s the multimodal (text, image, video, etc., complex data processing) technology secured by the game company. Technologies accumulated over time, such as NPC (Non-Player Character) that learns the world view through AI and communicates with users like a human, automatic translation, and voice synthesis, have shone.
Consider whether the physical world can be implemented with only text information. The surrounding environment must be recognized, and sufficient data must be available to create a good model. Our strength comes from here. Multimodal is important, and games are where technologies like 3D animation are comprehensively implemented. By researching the game domain, we have accumulated experience in this field and built a lot of high-quality training data. We have developed the capability to release multimodal commercial models suitable for the domestic industrial field, using high-quality 3D animation, sound, and text-to-speech (TTS) data accumulated over a long period as core assets.”
Provided by NC AI
What was the background for securing the ‘From Scratch’ technology to build the model directly from the very beginning?
“Having the capability of ‘From Scratch’ influenced this result. It is necessary to have internal capabilities to fix and improve errors when they occur. Currently, there are few companies in Korea with experience and technology in ‘From Scratch’. There are limitations to modifying already released models. We decided to start ‘From Scratch’ from 2022, even though it was costly and difficult when ChatGPT was released. As a result, we were able to sequentially release our own LLM ‘Barco 1.0’ and ‘2.0’.”
Are there specific collaborative projects requested by the Grand Consortium partners?
“‘Digital Twin’ is representative. It involves applying digital twins to fields where it is risky and costly to do in the real world. For example, in the case of robots moving cars, the underbody of each car is different. Testing each one incurs high costs. By implementing digital twins with our ultimate multimodal foundation model, we can reduce related costs and streamline the system.”
What manufacturing sectors are you targeting?
“We are targeting fields that use robotic arms. HL Robotics is also participating in our consortium. We need to start realizing AX in manufacturing sectors, starting with steel and secondary batteries. Ultimately, we are also aiming for ‘Physical AI’.”
Big tech companies in the US and China are engaging in fierce AI hegemony competition with large-scale investments. What are our unique strengths compared to overseas big tech?
“Big tech companies emphasize their strengths mainly through benchmarking. However, to be well applied in actual industries, a lot of industry data needs to be incorporated. This requires the work of directly building and inputting data by receiving expert opinions.
To create a model suitable for industrial AX, we also need to accumulate data from manufacturing, public, and media sectors. As a manufacturing powerhouse, Korea has a lot of high-quality data, which I believe is advantageous compared to big tech.”
ⓒ dongA.com. All rights reserved. Reproduction, redistribution, or use for AI training prohibited.
Popular News
