LG AI Research President Bae Kyunghoon (File photo by LG Corp.)
LG Group, South Korea’s fourth-biggest conglomerate, on Wednesday introduced the country’s first open-sourced artificial intelligence large language model (AI LLM) to compete with such technology of Big Tech names such as Meta Platforms Inc. and Google.
LG AI Research, the institute of the group’s holding company LG Corp., unveiled Exaone 3.0, an improved version of the predecessor showcased in July last year. The group revealed the first edition of Exaone in December 2021.
“Its performance is as strong as other open-sourced AI models such as Meta’s Llama and Google’s Gemma,” said an LG group official. “We succeeded in improving performance and reducing its size to three-hundredths compared to the initial model by focusing on lightweight and optimization technology.”
Exaone 3.0 has 7.8 billion parameters for learning on almost 8 trillion tokens. That compared with Google’s Gemma 2, which has 9 billion parameters on 8 trillion tokens.
The LG’s AI model ranked No. 1 in 13 benchmark areas, including real-world use cases, coding and math, demonstrating its global competitiveness when compared to other global open-source AI with similar size such as the Llama 3.1 and the Gemma 2, according to the South Korean group.
OPEN SOURCE AI
LG Group decided to open source the lightweight version of Exaone 3.0 to dominate the AI ecosystem by allowing industries to actively use the generative AI, industry sources in Seoul said.
Global Big Tech companies such as Microsoft Corp., Meta and Google rushed to develop lightweight AI models, which grew more popular due to better performances despite fewer parameters than those of other LLMs.
“LG does not need to depend on Nvidia’s AI semiconductor as it succeeded in developing a lightweight model,” said an electronics industry source in Seoul.
“The more AI engineers and students use the open-sourced Exaone 3.0, the wider LG’s AI-related businesses will be.”
GLOBAL PARTNERSHIPS
The group plans to introduce the latest AI LLM to the products and services of its affiliates such as the world’s leading home appliance maker LG Electronics Inc. and the conglomerate’s mobile carrier LG Uplus Corp.
Those units are poised to optimize Exaone 3.0 based on their products and services to speed up innovation.
LG Group is actively in talks for global partnerships, sources said.
“We aim to vitalize the AI research ecosystem and improve the national AI competitiveness by unveiling the country’s first open-sourced homegrown AI model,” said LG AI Research President Bae Kyunghoon.
CHATEXAONE
The group also unveiled ChatExaone, an enterprise AI agent based on the latest AI LLM and started its open beta service for affiliates’ staff.
ChatExaone provides various functions to enhance work productivity, including real-time web-based Q&A, document and image-based Q&A, coding, and database management, LG Group said.
That will allow group employees to utilize AI for their jobs such as searching, summarizing, translating, data analyzing, coding and others, the conglomerate added.
ChatExaone adopted the retrieval-augmented generation (RAG) technology, which utilizes real-time web search results to understand the context of the user’s instruction and provide answers that reflect the latest information, LG Group said.
By Chae-Yeon Kim
why29@hankyung.com
Jongwoo Cheon edited this article.
