[Edaily Reporter Kim Hyun-ah ] Ahead of the third evaluation of its proprietary AI foundation model (Dokpamo), which is scheduled to run through January of next year, SKTelecom(017670)is partnering with domestic mathematics and artificial intelligence (AI) researchers to enhance the inference capabilities of Dokpamo’s “A.X K2.”

The goal is to enhance the model’s ability to solve problems using specialized knowledge and verify the validity of its solutions through research on challenging mathematical problems. SK Telecom plans to incorporate the results of this mathematical research into the model’s refinement and utilize them to develop Agent AI capable of performing complex tasks.

SKTelecom announced on the 22nd that it will hold a technical exchange meeting with researchers from Seoul National University and the Korea Advanced Institute of Science and Technology (KAIST) at its headquarters in Euljiro, Seoul, to discuss ways to improve A.X K2’s mathematical reasoning performance and solve difficult problems. Domestic experts, including mathematicians from the Institute for Basic Science (IBS), will also participate in the joint research.

Strengthening A.X K2’s Reasoning Capabilities by Solving Mathematical Problems

This research aims to enhance

A.X K2’s reasoning capabilities

through the process of solving

mathematical problems

. Mathematics is a field that requires analyzing complex conditions, linking multiple logical steps to derive conclusions, and then verifying their validity. The research team plans to incorporate mathematicians’ specialized knowledge into the model’s training and apply it to solving real-world mathematical problems to elevate A.X K2’s reasoning capabilities.

Experts in specialized fields such as probability theory and statistical physics, discrete mathematics and graph theory, and topology and geometric group theory are participating in the research. Researchers from Seoul National University and KAIST are also exploring ways to utilize large language models (LLMs) in mathematical research.

This aligns with the trend among global AI companies to test and advance their models’ reasoning capabilities through mathematical research. On the 6th of this month, OpenAI unveiled new mathematical research findings derived by its AI model. Anthropic also announced last August that its AI model, Claude, had improved upon existing research results on a problem related to the Riemann Hypothesis, and that the findings had undergone peer review by mathematicians.

As the competitive edge of AI expands from simple knowledge acquisition or generating correct answers to the ability to solve complex problems and verify results, mathematical research is gaining attention as both a testing ground and a research tool for model advancement.

Mathematical Reasoning Capabilities to Drive Agent AI Advancement

SKTelecom plans to leverage the reasoning capabilities enhanced through mathematical research to enable

Agent AI

to perform complex tasks. The company aims to improve the Agent AI’s ability to independently verify the validity of its judgments and outcomes at each stage as it devises solutions to complex problems and carries out multi-step tasks.

In the long term, the company also plans to establish a research foundation for “Recursive Self-Improvement (RSI),” where AI improves its own performance and then uses those enhanced capabilities to drive the next stage of development. To achieve this, the AI must be able to judge not only the results it generates but also the validity of the experiments and verification processes it designs on its own.

SKTelecom aims to enhance A.X K2’s reasoning performance through joint research with experts in mathematics, and to link this to the development of practical problem-solving capabilities and the advancement of Agent AI.

Kim Tae-yoon, Head of Foundation Models at SKTelecom, stated, “We are pursuing industry-academia collaboration in the field of mathematics to enhance the problem-solving capabilities of our proprietary foundation model,” adding, “We will advance the capabilities of Agent AI through joint research with expert researchers.”