KT bets big on large-scale AI language models

Industry

Published: 11 Sep. 2023, 18:35 Updated: 11 Sep. 2023, 18:50

KT bets big on large-scale AI language models

KT CEO Kim Young-shub speaks to AI startup Qanda CEO Lee Yong-jae at the Mobile 360 Asia Pacific conference in Jung District, central Seoul, on Thursday. KT announced on Sunday that it formed strategic partnerships with Upstage and Qanda, investing a total of 20 billion won. [KT]

KT's latest investments and partnerships announced at the helm of newly-appointed CEO Kim Young-shub reflect the company's focus on large-scale artificial intelligence.

KT invested 10 billion won ($7.52 million) each in two local startups, Upstage and Qanda.

Upstage is known for its AI chatbot AskUp. Its 30-billion parameter large language model (LLM) ranked No. 1 on Hugging Face’s open LLM Leaderboard. Hugging Face acts as a hub for AI experts and enthusiasts, and the LLM leaderboard grades the models based on standards such as the model’s reasoning challenge, common sense interference, contextual understanding and factual accuracy.

Education technology startup Qanda runs online tutoring and problem-solving services utilizing its software such as optical character recognition, which converts documents or text images into machine-readable texts.

“KT aims to vitalize the local LLM ecosystem through these partnerships,” said Song Jae-ho, vice president of KT’s AI/DX Convergence Division in a statement.

Mobile carriers like KT have been looking to expand its business beyond telecommunication, accelerating its efforts to develop new growth engines such as AI, big data and cloud services.

KT is scheduled to release its large-scale model Mi:dm as early as October. If the launch is successful, KT will become a handful of companies with its own homegrown large language model.

Upstage CEO Kim Sung-hoon speaks at a session to talk about AI at the Mobile 360 Asia Pacific conference in Westin Josun Hotel in central Seoul on Thursday. [KT]

KT has been making investments worth billions of won in AI companies to perfect its model. The company invested 30 billion won in AI chip startup Rebellions in July 2022, and 15 billion won in AI software company Moreh this July.

KT’s recent deals in Upstage and Qanda show the company's willingness to strengthen sectors in business-to-business (B2B) and business-to-consumer (B2C) markets.

KT elaborated that with Upstage, it will develop large language models tailored to companies, probing into the B2B market in AI. With Qanda, KT will collaborate in the B2C realm to develop language models tailored to education-related services.

The global race for AI supremacy has now spurred another race among companies trying to reap profits from the large language models. Another mobile carrier SK Telecom is working to expand its AI-related services through alliances with foreign partners.

The company formed the “Global Telco AI Alliance” with major telecommunications firms from Germany, the United Arab Emirates and Singapore in July. It also created “K-AI Alliance” with local AI startups in June to strengthen its AI business.

From left, Jonathan Abrahamson, head of product management for digital at Deutsche Telekom, Khalifa Al Shamsi, CEO of e& life, and Anna Yip, deputy CEO of Singtel, and SK Telecom’s CEO Ryu Young-sang pose for a photo at the Global Telco AI Alliance CEO Summit meeting on July 27 at the Grand Walkerhill Seoul hotel, eastern Seoul. [SKT]

The company invested $100 million in San Francisco-based AI startup Anthropic in August, as well as in a local startup called Conan Technology in October 2022. Through the investment, SKT is aiming to attract the B2B market to fuse its in-house AI service “A” with other AI systems.

Last month, Naver revealed its new large-scale AI service HyperClova X, and also launched its B2B service “Neurcloud for HyerClova X.” The service provides its AI system to companies that cannot develop its own models, and ensures data protection as the cloud server will be placed inside the client's building. Naver is said to be in talks with various large corporates, ultimately aiming to glean some profits from its much-invested model.

KT’s business model for its B2B service in the AI sector is expected to be outlined within the year, based on the collaboration of Mi:dm and data collected from KT Cloud, the company’s cloud computing subsidiary.

“We are preparing to roll out new hyperscale AI services for companies in public and financial sectors that offer greater data security and protection,” KT said.

BY YU SUNG-KUK [lee.jaelim@joongang.co.kr]