SK Telecom works with Seoul National University, KAIST to boost A.X K2 maths reasoning
In short
SK Telecom said on 22 October that it is working with Seoul National University and KAIST researchers to strengthen the maths reasoning of its own AI foundation model A.X K2. The group plans to put mathematicians' expertise into the model's training and then try to solve hard maths problems with it. SK Telecom also plans to use the reasoning ability for agentic AI.
Read the full story 2 min read
SK Telecom said on 22 October that it is working with Seoul National University and KAIST researchers to raise the maths reasoning ability of A.X K2 (에이닷엑스 케이투), its own AI foundation model. The company said the plan is to teach the model specialised maths knowledge and then use it in research on hard maths problems. [ 1 , 2 , 3 , 4 ]
SK Telecom is holding a technical exchange meeting at its headquarters in Euljiro, Seoul, with professors and researchers from Seoul National University and KAIST. They are to discuss research on hard maths problems using A.X K2 and ways to upgrade the model. Newsis and Asia Economy said the attendees include Seoul National University mathematical sciences professors Seo Inseok (서인석), Byun Seongsu (변성수) and Hong Youngjoon (홍영준), KAIST computer science professor Park Noseong (박노성) and student researchers. [ 1 , 2 , 3 , 4 ]
In mathematics, Seo Inseok works on probability theory and statistical physics. Lee Joonkyung (이준경) of Godeung Gwahagwon (고등과학원), who works on discrete mathematics and graph theory, and Kim Sanghyun (김상현) of the same institute, who works on low-dimensional topology and geometric group theory, also take part, according to Newsis, Asia Economy and Seoul Economic Daily. On large language models (LLMs) and methods for using them in maths research, Hong, Lee Dabin (이다빈), Oh Byeongkwon (오병권), Byun and Park are to cooperate. [ 1 , 2 , 4 ]
The researchers plan to reflect mathematicians' expertise in the model's training and to study how to use LLMs in real maths research. They also plan to make A.X K2 learn each sub-field of mathematics in greater depth, supplying knowledge and reasoning that earlier models did not sufficiently handle. Using the upgraded model, they plan to take on hard problems in each field. [ 1 , 2 , 3 , 4 ]
SK Telecom plans to use the maths reasoning ability to upgrade agentic AI, raising the ability to check whether each judgment and the final result are valid when an AI plans and carries out multi-step work. Newsis and Seoul Economic Daily linked this to research on recursive self-improvement, in which AI improves its own performance, saying it requires judging whether the AI's own experiments and verification methods are mathematically valid. The company's long-term goal is maths ability comparable to top global AI models and an AI agent that learns what it needs and develops on its own. [ 1 , 2 , 4 ]
Newsis and Seoul Economic Daily noted that global AI companies are also using maths research. They said OpenAI disclosed new maths research results from an internal AI model on 6 October, and that Anthropic said in August that its Claude had improved an existing result related to the Riemann hypothesis, which mathematicians reviewed. [ 1 , 4 ]
Kim Taeyoon (김태윤), who heads foundation models at SK Telecom, said: "We are actively pursuing industry-academia cooperation in mathematics to raise the real problem-solving ability of our own foundation model." He added that the company will upgrade its agentic AI capability through joint research with specialist researchers. [ 1 , 2 , 3 ]
Why it matters
The company presents the project as a way to improve its own foundation model's ability to solve real problems and to build agentic AI. The documents give no results yet, only plans.
Key facts
- SK Telecom said on 22 October it is working with Seoul National University and KAIST researchers on its foundation model A.X K2. [ 1 , 2 , 3 , 4 ]
- SK Telecom is holding a technical exchange meeting at its Euljiro headquarters in Seoul to discuss research on hard maths problems using A.X K2 and ways to upgrade the model. [ 1 , 2 , 3 , 4 ]
- Researchers plan to reflect mathematicians' expertise in the model's training and to study methods for using large language models in real maths research. [ 1 , 2 , 3 ]
- The aim is to cover specialised knowledge and reasoning that earlier models did not sufficiently handle, then take on hard problems in each maths field. [ 1 , 2 , 3 , 4 ]
- SK Telecom plans to use the maths reasoning ability to upgrade agentic AI. [ 1 , 2 , 4 ]
- Kim Taeyoon (김태윤), who heads foundation models at SK Telecom, said the company is actively pursuing industry-academia cooperation in maths. [ 1 , 2 , 3 ]
Confirmed by several sources
- SK Telecom said it is cooperating with Seoul National University and KAIST researchers to raise the reasoning ability of A.X K2 and to take on hard maths problems. [ 1 , 2 , 3 , 4 ]
- The meeting was held at SK Telecom's headquarters in Euljiro, Seoul, on 22 October. [ 1 , 2 , 3 , 4 ]
- The researchers plan to reflect mathematicians' expertise in model training and to study how large language models can be used in real maths research. [ 1 , 2 , 3 , 4 ]
Still unclear
- Which specific maths problems the project will try to solve The documents speak only of hard problems in each field and name none.
- Timeline and expected results of the project No document gives a schedule, targets or results.
- Link to recursive self-improvement research Only two of the four outlets describe this link, as an aim of the reasoning and verification ability.