LG Group’s artificial intelligence think tank LG AI Research on Wednesday unveiled the upgraded version of its hyperscale AI multimodal Exaone that was designed to learn expertise data and use it for research and development and marketing across industry sectors.
The Exaone 2.0 is the first upgrade of the AI language model that launched in December 2021. It understands both language and images simultaneously to generate image-to-text and vice versa. The latest hyper AI has learned about 45 million specialized documents, including patents and papers, and 350 million images that were secured through partnerships.
“LG is the only company in Korea that has commercialized both a bilingual model and a bidirectional multimodal model. We’ve developed a top 1 percent AI expert that understands and discovers the knowledge of the world,” LG AI Research chief Bae Kyung-hoon said during a press event in Seoul.
The group looks to create differentiated customer value from other generative AIs by accumulating successful practical cases through cooperation with its domestic and overseas partners, the AI think tank chief said.
While global big tech firms rush to get ahead in the race to launch their generative AIs, LG decided to nurture Exaone as a top 1 percent AI expert that can be used in almost all sectors such as manufacturing, research, finance and medicine. Exaone refers to “expert AI for everyone.”
Lee Hwa-young, a leader of AI business development unit at LG AI Research, said, “Based on professionalism and credibility, there has been no successful case in actual industrial sites and commercialization around the globe. … The most important task (for us) is to make a success story by meaningfully using our advanced AI.”
Since most of the current expertise data is in English, Exaone 2.0 was developed as a bilingual model in both English and Korean. The amount of training data the new version is able to process is over four times more than the previous model. It has also reduced inference processing time by 25 percent and memory usage by 70 percent, resulting in a 78 percent cost reduction. The image generation quality also reduced the inference processing time by 83 percent and cost by 66 percent, Bae explained.
LG also introduced three professional platforms for Exaone: Universe, Discovery and Atelier.
Exaone Universe is a chatbot service for professionals. The closed beta service is set to open to some LG AI researchers and partner universities on July 31 and the official service will be available to all researchers at the group.
Exaone Discovery is a platform for the development of new materials, substances and drugs to advance the chemical and bio sectors. Researchers at its affiliated firms will be able to use the platform in the fourth quarter, the center said.
Exaone Atelier, which helps with creative ideas, is a multimodal AI platform that expresses images in language and visualizes language as images. The service will be launched for professional designers inside and outside the group in the third quarter. While Exaone is a business-to-business service, part of Atelier will be available to the general public, Bae said.
The research center said it will take the lead in popularizing generative AI technology and develop it into a competitive AI company globally.
Regarding Exaone’s market share target, the center chief said it is still "too early to predict the market size."
The firm believes they could benefit from a first-mover advantage if it creates a strategic collaboration relationship with a top-tier partner in the market.
After establishing the AI research hub in 2020, LG announced a plan last year to invest 3.6 trillion won ($2.84 billion) in R&D in the AI and data field for the next five years to preoccupy future technologies and recruit talent. Since Koo Kwang-mo took office as the group chairman in 2018, LG has sought to nurture AI as a future growth engine.