SK Telecom eyes June launch for its telco-focused LLM

The Korean operator, keen to ensure its generative AI (GenAI) is in tune with the dynamics of the local market, is making use of not one but three LLMs. One has been developed in-house, and is called 'A.X', while the other two are OpenAI's GPT-4, and Anthropic's Claude.

These have been fed a fulsome helping of telco-specific training data – in Korean, naturally – that includes stuff like tariff plans, subsidies, loyalty schemes and such. They have also been trained on SKT's internal guidelines and its ethical AI values. They have even been trained to identify foul and abusive language.

The idea is that SKT will make use of whichever LLM is deemed most equipped to handle a particular query. So, one might fulfil the role of a customer care assistant, while another might be used for network operations, and another for office-based admin work, for example.

"It will not be easy to solve the various problems and provide the services that telecommunication companies are trying to provide with one general-purpose LLM," said Eric Davis, head of AI tech collaboration at SKT, in a statement translated from Korean.

"SKT's unique multi-LLM strategy is to create various telco LLMs...and allow them to be selected and used according to the situation," he said.

LLM development forms part of the base layer of SKT's AI pyramid strategy. This particular layer is all about AI infrastructure, and in addition to LLMs, also includes data centres for hosting AI applications, and semiconductors – like SKT's forthcoming X330 – designed specifically to handle AI workloads.

The focus turns towards application development in the middle layer of the pyramid, which SKT refers to as AI Transformation (AIX). It plans to work up a range of AI-powered solutions to overhaul its internal operations and business processes first and foremost, and then leverage this experience as the foundation for a suite of offerings to enterprise clients in various verticals.

This part of the plan is also starting to come together, with the unveiling of SKT's Intelligence Platform. SKT is starting close to home with this one – pitching it to telcos that want to efficiently build and develop GenAI applications.

SKT describes it as an 'enterprise AI development and operation package,' and lists call centre bots, chatbots, sales assistants, infrastructure assistants and in-house business innovation among the range of possible use cases.

"We expect that Korean and global telecommunication companies – as well as companies with similar business characteristics, such as consulting work – will be able to reduce the cost and time required to develop a large platform through the Telco LLM Intelligence Platform," SKT said.

The uppermost layer of the pyramid is concerned with various customer-facing services, like its A. (pronounced 'A dot') chatbot, among others, which will all be powered by that aforementioned multi-LLM tech.

If all goes according to plan, SKT aims to generate $18.5 billion in AI revenue by 2028. It's a bold ambition given that research firm and sister company Omdia predicts that the entire global GenAI market will be worth $58.5 billion by then.

Somewhat hopeful targets aside, it's impressive to see that SKT has – in a relatively short amount of time – charted a course towards the top of the GenAI pyramid.