"It is good that there is a national will not to miss the timing to build sovereign artificial intelligence (AI) in Korea. Naver is the pioneer of sovereign AI."
Seong Nak-ho, head of hyperscale technology at Naver Cloud, responded this way when asked about his feelings after being selected as one of the consortia to embark on the Korea-type sovereign AI development project. Sovereign AI refers to AI technology created solely with domestic technology.
Seong emphasized that no consortium can create 'AI that anyone in the nation can use' as well as the Naver Cloud team. Seong was met on the 12th at Naver's Green Factory in Bundang-gu, Seongnam.
The biggest goal of the Naver Cloud consortium is to 'research and development of foundational models for expanding AI accessibility for all citizens.' The omni foundation refers to a single model capable of integrated understanding and generation of heterogeneous data such as text, images, audio, and video. Organizations participating in the consortium include Naver, Twelve Labs, Seoul National University, Pohang University of Science and Technology (POSTECH), Korea University, and Hanyang University.
Seong highlighted that the differentiation factor lies in the 'full-stack' capabilities that encompass everything from AI model development to cloud and data center operations, and business-to-consumer (B2C) and business-to-business (B2B) services. He explained that when developing generative AI, the two most challenging issues to address are copyright and security. Seong noted, "It is difficult to find a company in the country that has dealt with both problems. Naver has already accumulated experience in developing and operating large-scale AI models, but other teams have hardly had any experience in actually creating and distributing models of this scale."
The consortium aims to establish a foundation that can extend to physical AI, including robotics, autonomous driving, and navigation, through the 'omni foundation model.' To this end, the video understanding AI company Twelve Labs has joined as a data lead partner. In addition, they collaborated with academia to bring in talents lacking access to graphics processing unit (GPU) opportunities for the project. Seong stated, "I believe that the division of roles among participating corporations in our consortium is the cleanest."
Seong stated that ultimately, various industry-specific customized models derived from a foundation model should emerge, defining it as the 'AI Plus X' model. He also mentioned plans to establish a nationwide AI service platform that would create an ecosystem where users can easily produce and distribute AI specialized in various fields, such as manufacturing, education, and public services, at a low cost. In the long term, the plan is to export platforms and solutions to global markets, particularly the Middle East and Southeast Asia, based on the data and technology secured domestically.
However, Seong stated that rather than surpassing the globally leading frontier models, he aims to secure the competitiveness of domestic models in terms of cost efficiency and specialized areas. The government has instructed that the project's goal is to achieve 95% of the performance of the latest global frontier-level AI models, and some teams have indicated that they will exceed this figure.
Seong acknowledged that there were criticisms regarding a lack of ambition in their consortium during the evaluation process, but he remarked, "It must be realistic." He continued, "Some are talking about creating models that exceed 100%, 105% of global models, but that is impossible," adding, "Rather than creating something novel, it is much more realistic and efficient to optimize it for domestic conditions and resources."
Seong asserted that government support is essential for securing data and copyright. He stated, "The government should include public data and broadcast materials in the scope of permitted training," and emphasized, "Copyright is currently in a gray area. No one has disclosed how most image generation models were trained." He added, "The AI data issue is quite sensitive and complex. It would be great if the government could step in to clarify."
Seong also expressed regret over not receiving direct support for GPUs in this project. He said, "It is a fact that being excluded from support simply because we already have GPU infrastructure as a company that started AI first is essentially reverse discrimination." He added, "If the government says, 'Don't worry about GPUs' but effectively excludes support, it may create constraints for business initiatives," and requested, "Please reflect the GPU possession status in the evaluation regardless."
Seong stated, "This project is not simply a technology contest but a test of national AI competitiveness," expressing his ambition, "Naver Cloud will go all the way to achieve practical usability and diffusion of the ecosystem."