Upstage launches alliance for AI model based in Hangeul

Upstage creates an alliance devoted to developing a Hangeul-based AI model. (Upstage)

Korean artificial intelligence startup Upstage said Monday that it is launching an industrywide alliance that is devoted to developing a large-scale language model that specializes in Hangeul, or the Korean alphabet.

The 1T Club -- meaning "1 trillion token club" -- alliance brings together private firms and various organizations to provide Korean-language data totaling over 100 million words.

A token refers to the basic unit of text or code used by an AI-powered language model to process and generate language.

Through the partnership, the company said, it aims to address the problem of a lack of Korean-language data and to develop a language model that can capture uniquely Korean cultural aspects.

Earlier this month, Upstage garnered global attention by launching the latest version of its own language model, which surpassed the performance of OpenAI's popular GPT-3.5.

Upstage’s language model achieved an average score of 72.3 points on the HuggingFace Open LLM Leaderboard, a global benchmark for evaluating open-source language models.

Through the 1T Club, the company said it also plans to establish an ecosystem that benefits both data providers and model developers. Especially under its profit sharing scheme, each partner is expected to receive earnings proportional to the volume of data contributed.

"Through the 1T Club, we aim to safeguard the interests of data providers while developing a language model that captures the essence of Korean culture,” Upstage CEO Kim Seong-hoon said.

“We will ensure that all domestic companies can reap the benefits of AI advancement to the fullest extent."

By Song Jung-hyun(junghyun792@heraldcorp.com)