SK Telecom (SKT) has announced a "multiple Large Language Model (LLM)" strategy to target the corporate and public "generative AI market."
On August 23, SKT announced that it is making technical and business preparations by combining A.LLM, which it has developed in-house, with Anthropic's LLM, which has strengths in ethical answers and large text input, and Konan LLM, which is rich in Korean data.
Currently, the Korean LLM market is dominated by OpenAI, and many companies are preparing their own LLM-based businesses. In addition to its own LLM, SKT has revealed plans to promote a "multi-LLM" business, including LLMs from other companies recognized in the industry.
First of all, in terms of self-strength, the company announced that it launched the Korean-based A.LLM's Foundation Model for Enterprise in August and is working on training and optimizing the model for customers.
LLM Foundation Model is an AI software with basic functions that can perform document summarization, document generation, and Q&A. In order for the basic functions to work properly, the model needs to be trained with real data from customers who need it and tuned in the direction the customer wants.
In terms of cooperation, SKT also plans to utilize Konan LLM, which has been expanding cooperation in various fields since its equity investment.
In the case of Anthropic's LLM, which announced its equity investment on August 13, the companies are first discussing how to combine LLM features with SKT's major B2B services. It is expected that actual service provision will be possible in the fourth quarter after the function integration work is completed.
AI services, including LLM, require data for pre-learning and GPU server infrastructure for stable service provision to enable sophisticated service development with reduced development time.
SKT is equipped with a large-scale GPU system, including the country's top supercomputer, which enables stable pre-training to meet customer needs.
In order to combine the secured LLM with a strong infrastructure and supply it in an optimized form for each industry, it is essential to collaborate with operators who have extensive experience in building and operating at actual customer sites.
In recent years, the market for LLMs has been shifting to the most cost-effective way to provide the specific services needed by customers in each industry.
SKT's LLM services in the enterprise sector aim to provide "industry-specific LLM services" that match each LLM service to the field where each LLM can best showcase its unique characteristics.
For example, in the case of A.LLM, there are service advantages for telecommunications companies such as conversations and chatbots due to the nature of the data and services learned, and in the case of Anthropic's LLM, the ethical principles that AI should follow have been learned in advance, so it is possible to obtain answers that meet the ethical principles set by humans.
Through this, SKT plans to preempt the market by focusing on A.LLM in the Korean-based built-in LLM market and utilize Anthropic's LLM in the cloud-based LLM market, where competition centered on super-sized AI is expected.
Meanwhile, when combined with SKT's big data analytics business, it is expected that workers in the manufacturing industry will be able to improve their productivity by reducing production time through LLM AI technology.
