[MWC26] SKT CTO Sukgeun Jung: "Effective AI Models Matter... Exaone Is Not the Best Model"

"LLM Evaluation Standards Should Be Diverse... Not Just One Criterion"
"We Will Develop Models That Can Be Effectively Used in Industrial Settings"

The head of SK Telecom's Artificial Intelligence (AI) division emphasized the importance of developing AI models that can be applied in real industrial settings. He suggested that, rather than focusing solely on benchmark-based evaluations, it is crucial to develop models that can deliver practical performance in everyday life and industry.

At a press conference held on March 3 (local time) in Barcelona, Spain, where the Mobile World Congress (MWC) 2026 is taking place, Sukgeun Jung, Chief Technology Officer (CTO) and Head of the AI Corporate Independent Company (CIC) at SK Telecom, stated, "The Ministry of Science and ICT said that LG AI Research's 'Exaone' is the best, but I don't think it is the most suitable model for real-life and industrial use."

Sukgeun Jung, Chief Technology Officer (CTO) and Head of the AI Corporate Independent Company (CIC) at SK Telecom, spoke at a press conference held on the 3rd (local time) in Barcelona, Spain, where Mobile World Congress (MWC) 2026 took place. Provided by SK Telecom

In the first evaluation of the government's independent AI foundation model project announced in January, LG AI Research's Exaone ranked first in the overall score, which was primarily based on benchmark assessments. However, the specific scores and rankings were not disclosed.

CTO Jung commented on the current AI model evaluation system, which is centered around quantifiable performance metrics such as benchmarks, saying, "There are various standards for evaluating the performance of large language models (LLMs). At present, we use benchmark scores, but just as there are different types of university admissions in Korea, it is not appropriate to evaluate AI intelligence based on a single criterion." He stressed the need for a broader and more diverse perspective in evaluating AI models.

He added, "SK Telecom intends to focus on creating models that can be effectively used in real industrial environments," and said, "I hope that starting from the second phase of evaluations, the government will also incorporate such considerations into its assessment criteria."

He also noted that, in competition with AI models developed by global big tech companies, it is important to establish realistic strategies for performance. CTO Jung said, "It would be extremely difficult to achieve 100% of the level of models like Gemini, ChatGPT, or Claude, but aiming for around 95% is feasible. In particular, there are many problems in manufacturing environments that can be solved with models performing at about 95% of that level."

Regarding the controversy over the originality of AI models during the first phase of the independent foundation model project, he said, "AI itself is becoming a technology of strategic value," and added, "There is no doubt that we need to possess our own capabilities."

When asked about monetization of SK Telecom's AI service 'A.Dot,' he answered, "The key is to first identify use cases that will make customers willing to pay for A.Dot," and added, "We are considering whether the service can deliver performance worthy of a paid offering."

This content was produced with the assistance of AI translation services.