Launches Expert AI Alliance with 13 Partners
LG AI Research announced on Feb. 22 that the Expert AI Alliance has been launched by 13 domestic and foreign companies from various business fields.
The alliance is the first private AI alliance formed by leading domestic and foreign companies in the IT, financial, educational, medical, manufacturing, communications sectors for cooperation among heterogeneous industries.
The 13 companies include Google, Woori Bank, Shutterstock, Elsevier, EBS, the Korea University Medical Center, Hanyang University Hospital, VA Corporation, LG Electronics, LG Chem, LG Uplus and LG CNS. These companies plan to develop services through collaboration based on Exaone, LG’s "super-giant AI."
Super-large AI can think, learn, and make judgments like humans by learning large-scale data on its own based on computing infrastructure capable of large-capacity computation. Tilda, the first AI artist implemented via Exaone, presented costumes, which were created in collaboration with designer Park Yoon-hee, at New York Fashion Week on Feb. 14.
LG AI Research also demonstrated an advanced chatbot model that summarizes content to the level of professional consultants after talking naturally like humans, and deep document understanding (DDU), a technology jointly developed with LG Chem and Elsevier. This technology analyzes and accumulates data quickly and accurately by analyzing not only texts from specialized literature but visual elements such as formulas and tables.
The institute plans to provide various services so that partners will be able to easily and conveniently use ultra-large AI. It plans to release 39 application programming interfaces (APIs) in the first half of 2022 so that customized professional AI can be easily developed. The APIs can be used in eight areas -- Q&A, text classification, summaries, conversations, text generation, keyword extraction, image generation, and image captioning.
Furthermore, the institute plans to accelerate the popularization of super-giant AI by operating the service platform Exaone Playground in the second half of 2022.
The institute also unveiled Exaone-Tuning, a new technology that will ease the burdens of data security and AI development period for partners. Exaone-Tuning reduces the number of parameters currently used by 90 percent, making itself much lighter, faster in processing and nearly four times more accurate than the most widely used method (P-Tuning). The institute stressed that specialized professional AI services such as AI bank employees may come out if it make Exaone additionally learn financial terms.