LG Showcases ExaOne Deep at GTC 2025
Highlights "Inference Feature" with Live Demonstration
Plans to Announce Integrated Version 4.0 in May
"Thinking..."
At NVIDIA's annual developer conference GTC 2025 held in San Jose, California, USA, LG AI Research showcased Korea's first inference artificial intelligence (AI) 'ExaOne Deep' on the 18th (local time). Bae Kyunghoon, head of LG AI Research, personally demonstrated the AI, drawing the full attention of the attendees.
When Director Bae input a question and pressed the 'Inference' button, the screen immediately displayed the message 'Thinking.' ExaOne Deep understood the context and core of the question, then autonomously suggested keywords to derive the most appropriate answer. It then quickly gathered and compared various data to instantly reach an accurate conclusion. The attendees witnessed the AI's human-like thinking process in real time.
Baek Gung-hoon, head of LG AI Research, demonstrated Korea's first inference AI, 'Exaone Deep,' at the LG AI Research booth during 'GTC 2025' held in San Jose, USA, on the 18th (local time). Provided by LG
The Birth of Korea's Representative Autonomous AI
Director Bae stated, "ExaOne Deep goes beyond simply answering questions; it has the ability to set hypotheses on its own, logically verify them, and derive the optimal conclusion." He added, "We will nurture ExaOne as Korea's representative autonomous AI." At the event, LG AI Research demonstrated in real time how ExaOne not only searches global data through the 'General' menu but also finds information more precisely via menus categorized by purpose, such as academic papers and YouTube videos. Particularly, the feature that analyzes the intent of questions and recommends the most suitable keywords attracted high interest from the audience.
In the 'Global' menu, data could be searched within overseas sources; the 'Academic' menu allowed searching for papers, and the 'Video' menu enabled searches within YouTube. The 'General' menu allowed searching across all sources. Director Bae explained the reason for advancing the announcement timing, saying, "Many companies have released versions integrating reasoning models into LMMs this year. We believed that if we could not develop a world-class inference model, we would not be able to prove the performance of integrated models."
He added, "We wanted to internally verify that ExaOne can compete with ChatGPT to some extent and demonstrate its competitiveness. We revealed it at GTC to promote ExaOne's global competitiveness and expand the model."
The internally developed LMM, ChatExaOne, is currently used by 40% of all employees across LG affiliates. However, it is only used within the affiliates and is not monetized. Director Bae emphasized, "We are not currently considering linking it to profitability. One of our goals is to integrate it with internal corporate documents so that both domain experts and general office workers can use it."
LG AI Research plans to announce version 4.0 of ExaOne around May. He said, "Many companies are expected to release integrated versions around May to June. ExaOne is also scheduled to be announced within this year."
Baek Gung-hoon, head of LG AI Research, demonstrated Korea's first inference AI, ExaOne Deep, at the LG AI Research booth during GTC 2025 held in San Jose, USA, on the 18th (local time). Photo by Park Joon-yi
Baek Gung-hoon, head of LG AI Research, demonstrated Korea's first inference AI, ExaOne Deep, at the LG AI Research booth during GTC 2025 held in San Jose, USA, on the 18th (local time). Photo by Park Joon-yi
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
