SKT Telecom LLM Briefing on the 30th
Collaboration with OpenAI 'GPT' and Anthropic 'Claude'
Improved Telecom Operations Efficiency in Customer Service and Infrastructure Management
SK Telecom will launch a telecommunications service-specialized 'Telco Large Language Model (LLM)' trained on telecom services, products, and memberships in June. It is expected to enhance the operational efficiency of telecom companies, from customer center consultations to telecom infrastructure management.
On the 30th, SKT held a briefing session for the 'Telco LLM' at the Suprex Hall in T Tower, Jung-gu, Seoul, announcing that the Telco LLM will be introduced in June. The Telco LLM is a telecom industry-specialized LLM trained on telecom-specific terms such as 5G plans, T Membership, official subsidies, as well as internal guidelines like AI ethics values.
SKT collaborates with OpenAI, Entropic, and others to collect data, select structured and unstructured data, and train it on general-purpose models such as 'A.DotX', 'GPT', and 'Claude'. Then, telecom-related data is further trained through fine-tuning.
Afterwards, the Telco LLM undergoes reinforcement learning based on human evaluation to determine whether it provides useful answers, followed by model evaluation (benchmarking) on language ability, reasoning skills, and telecom-specialized task performance. If the Telco LLM scores low in consultation type selection, related data is further built and trained to advance the technology.
Eric Davis, SKT AI Tech Collaboration Manager, said, "It will not be easy to solve the services and problems telecom companies want with a single general-purpose LLM," adding, "SKT’s unique multi-LLM strategy is to create various Telco LLMs through fine-tuning and benchmarking tailored to telecom data and domain know-how, and to select and use them according to the situation."
SKT expects the Telco LLM to significantly improve telecom customer consultation tasks. While agents talk with customers, the LLM can assist by summarizing consultation content and suggesting solutions. Typically, handling one consultation call takes about 3 minutes for customer interaction and over 30 seconds for post-call processing.
The Claude-version Telco LLM also has the advantage of accurately understanding contextual nuances such as AI ethical principles, Korean neologisms, profanity, and threatening language.
It is also useful for telecom network infrastructure management. Since the Telco LLM provides situation-appropriate answers based on equipment manuals and response cases, it can reduce response time compared to operators searching for information manually.
Jung Min-young, SKT AI Platform Manager, said, "Telco LLM will enhance operational efficiency in various areas of telecom operations, including customer centers, infrastructure, marketing, distribution channels, as well as internal tasks like legal affairs and HR."
On this day, SKT also unveiled the 'Intelligence Platform,' which enables telecom companies to efficiently build and develop telecom-specialized multi-LLMs. This is an 'enterprise AI development and operation package' covering multi-LLM, multimodal, orchestration, and retrieval-augmented generation (RAG).
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
