[THE VIEW]Securing Powerful Korean Language AI Models Is Essential

SLMs with Economic Efficiency
Korean-Specialized Versions by Naver and Others
Shortage of Global-Scale Korean Data
Industry-Specific Applications Needed, Including Semiconductors

Artificial intelligence (AI) technology has rapidly advanced over the past few years, deeply permeating various industries. In particular, the emergence of Large Language Models (LLM) has opened new horizons in text generation, translation, and data analysis. However, the high-performance computing resources required to process and train vast amounts of data, along with the enormous energy consumption, have been criticized as environmentally unsustainable. At the same time, these factors act as barriers limiting access to this technology for certain companies and countries.

To address these issues, Small Language Models (SLM) have emerged. Compared to large-scale models, SLMs have significantly fewer parameters while providing efficiency specialized for specific tasks. They are gaining attention as technologies that reduce dependence on the cloud and can be processed on local devices, contributing to data privacy protection and energy savings.

In the United States, SLMs are being utilized in combination with large-scale models in hybrid forms. Google has developed small models that operate on mobile devices, reducing cloud dependency and enhancing real-time responsiveness. OpenAI is researching lightweight versions of ChatGPT. Europe is leveraging SLMs while complying with strong data protection regulations such as the European Union’s General Data Protection Regulation (GDPR). In China, under strong government AI investment, SLMs are being customized for industrial sectors such as smart manufacturing.

South Korea possesses unique strengths in adopting SLMs but also faces considerable challenges. With fast internet infrastructure and high technological adaptability, Korea has a favorable environment for AI technology development. Naver’s ‘HyperClova X’ is an SLM specialized for the Korean language, exhibiting higher language comprehension compared to global models. Kakao has also successfully established itself in the domestic market by developing customized chatbots and search services using small language models. However, Korea faces several constraints in SLM development. The biggest issue is the lack of Korean language data. SLMs require extensive data training to be optimized for a specific language, but the volume of Korean data used globally is relatively small. Additionally, Korean companies’ strategies for entering the global market are insufficient, often remaining confined to the domestic market, which risks falling behind in global competition.

Korea needs to learn more actively from global cases. It is necessary to establish Korea’s own strategies by referring to the U.S. hybrid solution cases and Europe’s regulation-friendly AI development approaches. Especially important is the application of SLMs tailored to industries such as semiconductors, smart manufacturing, and content, which are Korea’s key sectors. This approach can be similar to China’s method of maximizing efficiency by processing industrial data locally. At the same time, support for AI startups should be expanded to promote innovation based on SLM technology and provide networks for global expansion.

Economically efficient SLMs are reshaping the global AI ecosystem, and Korea must join this trend to strengthen its competitiveness. For this, a creative and strategic approach connecting global trends and local characteristics is essential. SLMs, though small, are powerful technologies that will play a crucial role in opening a new era of AI together with Korea and the world.

Yunseok Son, Professor at the University of Notre Dame, USA

This content was produced with the assistance of AI translation services.