KT teams up with KAIST to develop hyperscale AI model

August 18, 2021

KT Corp., a major South Korean telecom operator, said Wednesday it has joined hands with the country’s top science university KAIST to develop a hyperscale artificial intelligence (AI) language model.

The telecom operator plans to commercialize the model, which uses massive computing infrastructure for natural language processing, in the first half of next year, with hopes of powering its services, such as smart speakers and AI call centers.

The Electronics and Telecommunications Research Institute and Hanyang University, members of KT’s AI alliance, will also take part in the joint research project.

The collaboration aims to eventually develop an AI model with more than 200 billion parameters. Higher parameters usually lead to a model’s increased capacity. In comparison, U.S.-based OpenAI’s GPT-3 model, which can produce human-like text, contains 175 billion parameters.

KT joins a growing list of South Korean companies developing AI models.

LG Group said in May it will invest US$100 million in the next three years to develop a “super-giant AI” powered by large scale computing infrastructure.

In the same month, internet giant Naver Corp. unveiled its AI language model “HyperCLOVA,” which has 204 billion parameters.

This file photo, taken on Nov. 3, 2020, shows KT Corp.'s headquarters in central Seoul. (Yonhap)

This file photo, taken on Nov. 3, 2020, shows KT Corp.’s headquarters in central Seoul. (Yonhap)