Do not go gentle into that good night, old age should burn and rave at close of day; Rage, rage against the dying of the light, though wise men at their end know dark is right, because their words had forked no lightning they, do not go gentle into that good night.
SK Telecom to deploy proprietary AI model in steel, auto parts factories
An infographic showing SK Telecom's artificial intelligence foundation model deployed to partner companies. Courtesy of SK Telecom
By Jhoo Dong-chan
SK Telecom (SKT) has signed memorandums of understanding (MOU) with steel manufacturer KG Steel and automotive parts maker KONEC to pilot its proprietary artificial intelligence (AI) foundation model in their production facilities — marking the first deployment of the model in the manufacturing sector.
SKT has been developing the model, called A.X K1, since earlier this year.
The company began collecting process error reports, equipment manuals and operational log data from KG Steel and KONEC in April and has since completed a demo version of what it calls a "manufacturing-specialized AI agent" built on the model.
A.X K1 is a large language model with 519 billion parameters — the numerical variables that determine a model's behavior. The company said the model is structured so that only about 33 billion parameters activate during each inference, allowing it to run efficiently in industrial environments while retaining the reasoning capacity of a much larger system. Parameters are a standard measure of a language model's scale and complexity.
In the second half of the year, SKT and its two partners plan to deploy the demo agent in KG Steel's cold-rolling line at its Dangjin plant — which produces coated steel sheets — and in KONEC's die-casting and machining processes. KG Steel produces galvanized and cold-rolled steel sheet, while KONEC makes aluminum die-cast components for automotive drivetrains and is a supplier to Tesla and major Korean automakers.
The pilot arrangement calls for the two manufacturers to share real-time process data with SKT, which will use that feedback to improve the agent's performance and extend its capabilities. Data gathered during the pilot will also be used to train A.X K2, SKT's next-generation model currently under development.
SKT said manufacturing has been a difficult domain for AI adoption because shop-floor data is fragmented across departments and processes, and critical knowledge tends to remain with individual skilled workers rather than being systematically documented.
It also noted that manufacturing sites are often reluctant to adopt cloud-based AI because it requires sending sensitive process data to external servers. SKT said A.X K1 supports on-premise deployment, allowing the model to run inside a company's own infrastructure without data leaving the facility.
The manufacturing push follows an MOU the telecom company signed last month with the Ministry of National Defense to apply the same model in defense applications. SKT said it plans to expand into finance, public services and health care.
This article was published with the assistance of generative AI and edited by The Korea Times.