SK Telecom is set to tackle mathematical challenges using its proprietary AI foundation model, "A.X K2."
SKT announced that it will hold a technical exchange meeting on the morning of the 22nd at its Seoul Eulji-ro headquarters with faculty and researchers from Seoul National University and KAIST to discuss research on mathematical challenges utilizing A.X K2 and strategies for model advancement.
In the field of mathematics, experts are contributing their specialized research capabilities, including Seo In-seok (Prof.) from the Department of Mathematical Sciences at Seoul National University, who studies probability theory and statistical physics; Lee Jun-gyeong (Prof.) from the Department of Computational Science at the Institute for Basic Science, who studies discrete mathematics and graph theory; and Kim Sang-hyeon (Prof.) from the Department of Mathematics at the Institute for Basic Science, who studies low-dimensional topology and geometric group theory.
In the areas of large language models (LLMs) and mathematical research methodologies utilizing them, Hong Yeong-jun (Prof.) from the Department of Mathematical Sciences and the Interdisciplinary Program in Artificial Intelligence at Seoul National University; Lee Da-bin, Oh Byung-gwon, and Byeon Seong-su (Prof.) from the Department of Mathematical Sciences at Seoul National University; and Park No-seong (Prof.) from the Department of Computer Science at KAIST are collaborating. They plan to jointly research methodologies for incorporating mathematicians' expertise into model training and applying LLMs to actual mathematical research.
The research team will collaborate to enable A.X K2 to deeply learn each subfield of mathematics. They aim to supplement specialized knowledge and reasoning capabilities that existing models have not adequately addressed, and plan to challenge the resolution of mathematical challenges in various fields using the advanced model.
SKT will leverage the enhanced mathematical reasoning capabilities gained from this industry-academia collaboration for the advancement of agentic AI. The company plans to improve the ability to verify whether the decision-making processes and results are valid when an AI formulates a solution plan for complex problems and executes multi-stage tasks.
Reasoning and verification capabilities are also linked to AI self-improvement research. Recently, in the AI field, "Recursive Self-Improvement (RSI)" is being studied, where AI autonomously improves its own capabilities and utilizes those enhanced abilities for further improvements at subsequent stages. The ability to judge whether the results produced by an AI and the experiments and verification methods it designs on its own are mathematically valid is crucial.
In the long term, the goal is to secure mathematical capabilities comparable to top-tier global AI models and to realize AI agents that learn necessary knowledge and develop autonomously.
Kim Tae-yoon, head of foundation models at SKT, stated, "We are actively promoting industry-academia collaboration in mathematics to enhance the real-world problem-solving capabilities of our proprietary foundation model," adding, "Through joint research with specialized researchers, we will continue to advance our agentic AI capabilities."
in TECH
