[AI Revolution](123) '5.7 Billion Won Investment Secured' Trillion Labs "We Will Create an LLM Like a Korean"

Interview with Trillion Labs CEO Shin Jae-min
"Developing Korean LLM trained on Korean language data"
Raised 5.7 billion KRW investment from Kakao Ventures and others

"We will embed Korean from the development stage to create a large language model (LLM) that is like a Korean."

AI Revolution Interview - Shin Jaemin, CEO of Trillion Labs, is being interviewed by Asia Economy at the office in Gangnam-gu, Seoul. Photo by Jo Yongjun jun21@

Recently, Shin Jaemin, CEO of Trillion Labs, whom we met at a shared office in Gangnam, Seoul, described the Korean-based LLM currently under development this way. Founded in April this year, Trillion Labs recently secured $4.2 million (about 5.7 billion KRW) in pre-seed investment from Strong Ventures, Kakao Ventures, Base Investment, The Ventures, US-based Goodwater Capital, and BAM Ventures.

The reason Trillion Labs was able to attract investment was that its business model of creating a Korean LLM foundation model drew great interest. While existing commercialized Korean LLMs provide Korean by fine-tuning English models trained on English data, Trillion Labs aims to input Korean from the training stage. CEO Shin said, "We want to create an LLM that answers in Korean by inputting large-scale Korean data to build an AI that is smart like a Korean." The goal is to enhance the competitiveness of key industries where Korea has strengths, such as content and advanced manufacturing.

He cited "email writing tasks" as an example of a Korean-optimized LLM. CEO Shin explained, "If you write a business email using ChatGPT, the overall greetings and structure are formatted in the style of English-speaking cultures," adding, "For example, the last phrase might be translated literally as 'Thank you, Shin Jaemin,' following the English email format 'Sincerely, Shin Jaemin'." He continued, "In the case of an LLM trained in Korean, the last phrase ends like a real Korean would write it, such as 'Shin Jaemin Dream'." The Korean LLM model reflects Korean culture and sentiment in its responses. He said, "It's a very subtle difference but an example that shows whether it looks like it was written by a Korean or not."

AI Revolution Interview - Shin Jaemin, CEO of Trillion Labs, is being interviewed by Asia Economy at the office in Gangnam-gu, Seoul. Photo by Jo Yongjun jun21@

For data expansion, it also trains on English, Chinese, and Japanese data. CEO Shin said, "Since half of the world's information is in English, we inevitably have to include it," and added, "If we bring in Chinese and Japanese, which are part of the Han character cultural sphere, for example, universities (大學校) that use the same Chinese characters in the three countries are treated as the same entity, making training more efficient."

CEO Shin plans to actively utilize his accumulated experience and knowledge in developing the LLM. He majored in Computer Science for both undergraduate and master's degrees at Hong Kong University of Science and Technology, and later worked as an AI research intern on Amazon Alexa and as a core researcher in developing Naver's LLM, HyperCLOVA X. He has authored numerous papers on synthetic data generation, empathetic dialogue systems, and Korean natural language processing (NLP). CEO Shin particularly emphasized the use of synthetic data knowledge. He said, "For example, if you want to create an AI that supports lawyers, you would have to hire lawyers to create data, but the cost burden is high," adding, "Synthetic data can produce high-quality data based on a small amount of lawyer data."