SK电讯携手首尔大学、KAIST数学研究者强化自研AI模型A.X K2推理能力

SK Telecom is working with mathematics and artificial intelligence researchers at Seoul National University, KAIST and other domestic institutions to strengthen the mathematical reasoning of its proprietary AI foundation model. The company aims to improve the model's problem-solving ability through research on unsolved mathematical problems and, over the longer term, lay the groundwork for developing AI agents that learn and improve on their own.
SK Telecom said on the 22nd that it would hold a technology exchange session with professors and researchers from Seoul National University and KAIST at its headquarters in Euljiro, Seoul, to discuss research on unsolved mathematical problems using its proprietary AI foundation model A.X K2 and ways to upgrade the model.
Mathematics and AI specialists from Seoul National University, KAIST and the Korea Institute for Advanced Study are taking part in the collaboration. The mathematics research will be led by Seo In-seok, a Seoul National University professor specializing in probability theory and statistical physics; Lee Jun-kyung, a professor at the Korea Institute for Advanced Study working on discrete mathematics and graph theory; and Kim Sang-hyun, a professor at the same institute specializing in low-dimensional topology and geometric group theory. Seoul National University professors Hong Young-jun, Lee Da-bin, Oh Byung-kwon and Byun Sung-soo, along with KAIST professor Park No-sung, will study methods for applying large language models to mathematical research.
The researchers plan to incorporate specialized knowledge from each field of mathematics into the training of A.X K2 to strengthen its reasoning on advanced problems that existing models have not handled adequately. Building on that work, they will also attempt to solve actual unsolved mathematical problems.
Global AI companies have recently been using mathematical research to test the reasoning capabilities of their models. OpenAI on the 6th of this month released new mathematical research results produced by an internal AI model. Anthropic said in August that its AI model Claude had improved on existing research related to the Riemann hypothesis and that the work had been reviewed by mathematicians.
SK Telecom plans to apply the findings to the development of agentic AI. The key goal is to strengthen the AI's ability to verify on its own the validity of its judgments and results as it draws up plans and carries out multistep tasks to solve complex problems.
The company also intends to lay the foundation for research into recursive self-improvement (RSI), in which AI repeatedly enhances its own performance. The aim is to raise the model's reasoning and verification capabilities so that it can assess not only the results it produces but also the validity of the experiments and verification methods it designs itself.