SK Telecom has made waves in the tech sector with its selection as a core team for the ‘Proprietary AI Foundation Model’ project, an initiative spearheaded by the Ministry of Science and ICT (MSIT). This collaborative effort aims to create a robust full-stack artificial intelligence (AI) platform leveraging proprietary technology across semiconductors, models, data, and services, setting a strong precedent for local innovation in South Korea.
A cornerstone of this partnership with MSIT involves open sourcing the project’s capabilities, a move anticipated to empower local businesses and significantly enhance the country’s AI innovation landscape. The SK Telecom Consortium, comprising industry giants and research institutions, is committed to advancing proprietary AI. Their impressive track record includes publishing over 800 research papers, securing 736 patents, and launching upwards of 270 open-source projects.
Leading the core research for the MSIT project is Kim Tae-yoon, who oversees the Foundation Model Office at SK Telecom. He is joined by esteemed colleagues, including Professors Lee Kangwook and Dimitris Papailiopoulos from the University of Wisconsin-Madison. The consortium also features a variety of contributors, from gaming powerhouse Krafton to mobility AI firm 42dot, and chip manufacturer Rebellions, not to mention a host of eminent professors from Seoul National University and KAIST.
SK Telecom is gearing up to “embark on the development of next-generation large-scale AI models” through this initiative. The company plans to harness omni-modal technology, enabling it to process text, images, speech, and video simultaneously—because who doesn’t want their AI multitasking like a seasoned professional?
Since 2018, SK Telecom has been on a mission to develop its own large language model (LLM), known as A.X. This year marked the introduction of two iterations of A.X 4.0—both standard and light versions—crafted through extensive continual pre-training. The company has also created two distinct versions of A.X 3.1, initiated from scratch. These models boast performance on par with GPT-4o, demonstrating exceptional prowess in understanding the nuances of the Korean language.
To date, these A.X models have been trained on SK Telecom’s state-of-the-art TITAN supercomputer. Looking ahead, the SKT Consortium is set to provide substantial computing resources autonomously for advanced research and development (R&D). Moreover, consortium partner Rebellions will deploy its domestically developed neural processing units (NPUs) for optimizing high-performance, energy-efficient AI services.
The ultimate vision for this consortium is to pave the way for making AI agents accessible and practical for all Koreans. The focus spans a range of key sectors, including office work, manufacturing, automotive, gaming, and robotics—it’s clear that AI isn’t just a trend; it’s becoming part of the fabric of everyday life in South Korea.
In the words of Kim Jiwon, Head of AI Model Lab at SK Telecom:
With our proven technological capabilities and operational expertise, we will deliver the highest-quality, Korean-style, proprietary AI foundation model to empower AI for daily life in Korea.
What is the main goal of the SKT Consortium’s project with the MSIT?
The primary aim is to develop a proprietary artificial intelligence platform that promotes local innovation and enhances the South Korean AI ecosystem through open-source capabilities.
Who are some key figures involved in the research for the project?
Kim Tae-yoon leads the initiative, supported by notable researchers including Professors Lee Kangwook and Dimitris Papailiopoulos, along with representatives from various companies and universities.
What technologies will be emphasized in the development of new AI models?
The consortium plans to integrate omni-modal technology, enabling simultaneous processing of text, images, speech, and video data, positioning it at the forefront of AI advancements.
