Korea is using its own artificial intelligence large model to challenge the global AI landscape.

The Korea Herald's front page reported that Korean AI challengers are using their language models to challenge ChatGPT. LG, Naver, SKT, Upstage, and NC Al are leading in the autonomous AI competition by building models for Korean and context.

In a series of reports, Kakao's Kanana and NC AI's Varco are competing to take a place in the government's "autonomous AI-driven" initiative.

Five consortia led by LG AI Research, SKT, Naver, NC AI, and Upstage have been selected to participate in an initiative aimed at providing autonomous AI foundational models.

Although their models have hundreds of billions of parameters, compared to trillions of parameters in leading language models, the project aims to achieve at least 95% performance of the cutting-edge models (such as ChatGPT).

Ultimately, two final teams will be selected through a series of evaluations. LG AI Research's Exaone 4.0 was launched in July and quickly became the flagship competitor in South Korea's autonomous AI competition. It is one of only two Korean language models listed in the Artificial Intelligence Index (a renowned index that evaluates global AI models).

At launch, it ranked first among Korean models, fourth among open-weight models, and 11th overall in the global AI index. The system has two formats: a high-performance 32B model (representing 32 billion parameters) for compute-intensive tasks, and a lightweight 1.2B version optimized for device usage.

It also has visual analysis capabilities, enabling it to process unstructured data such as images, documents, and charts. In a specialized "chart question answering" benchmark test, which measures the accuracy of interpreting company documents and charts, Exaone achieved one of the highest scores globally.

Korea is strongly entering the global large language model competition, with enterprise groups and Korean AI startups launching AI systems tailored for Korean language and culture, aiming to challenge the dominance of OpenAI and Google.

LG's language model primarily focuses on enterprise needs, such as automating workflows, analyzing data, and supporting research in industries such as finance, pharmaceuticals, and manufacturing.

From LG's Exaone, Naver's HyperClova X, to Upstage's Solar Pro, and SK Telecom's A.X, The Korea Herald investigated the strength comparison of these participants and how other challengers are trying to carve out space. This hybrid language model, which combines reasoning functions, shows competitive performance compared to some frontier models that are several times larger. Among the five selected in the rapidly developing language model field in South Korea, LG AI Research's Exaone: a compact hybrid model that can compete with larger frontier models.

Original article: www.toutiao.com/article/1842200309850124/

Statement: The article represents the views of the author.