Upstage, Flitto join hands to build multilingual AI dataset for large language models

By Kim Joo-heon Posted : May 10, 2024, 14:26 Updated : May 10, 2024, 14:26
Getty Images Bank
SEOUL, May 10 (AJU PRESS) - Korea's AI startup Upstage has signed a memorandum of understanding with translation platform operator Flitto to create an artificial intelligence-based dataset that can handle multiple languages beyond English.  

The partnership inked Thursday is aimed at developing large language models (LLMs) in Japanese, Thai, and other languages, potentially filling the data gap left by major western tech firms' LLM development projects, Upstage said. Western projects are primarily focused on English, resulting in a deficiency of data in Asian languages.

Under the project, Upstage's LLM solution Solar will be integrated with Flitto's language data. Solar is a deep-learning platform capable of understanding and generating content. In August 2023, the solution claimed the top spot on an international large language model scoreboard run by Hugging Face, a New York City-based AI company. Solar has competed against some 500 models including GPT-3, a Chat GPT-based service.

Flitto currently offers translation services in 25 languages including English, Korean, Vietnamese and Thai. The company has worked with Seoul City to provide an AI translation service that engages in conversation with foreign tourists by displaying text on the screen at tourist information centers.

"Through this collaboration with Flitto, we will endeavor to elevate the sophistication of our data, enabling a broader global audience to engage with generative AI innovation," Upstage CEO Kim Seong-hoon said after the signing ceremony.

Copyright ⓒ Aju Press All rights reserved.

기사 이미지 확대 보기
닫기