LG, Google Collaborate to Build a 'Supermassive AI Ecosystem'

LG AI Research Institute Launches AI Consortium with 12 Domestic and International Companies
"Nurturing Top 1% AI Experts... Leading the Popularization of Super-large AI"

[Asia Economy Reporter Lee Hyeyoung] LG AI Research Institute has launched an AI consortium with 12 domestic and international companies, marking the beginning of a full-scale ecosystem expansion.

On the 22nd, LG AI Research Institute announced the launch of the 'Expert AI Alliance' with leading companies from the IT, finance, education, medical, manufacturing, and telecommunications sectors both domestically and abroad. This is the first private consortium formed for AI collaboration across different industries.

The consortium, led by LG AI Research Institute, includes founding members such as Google, Woori Bank, Shutterstock, Elsevier, EBS, Korea University Medical Center, Hanyang University Hospital, V Corporation, and LG Electronics, LG Chem, LG Uplus, and LG CNS.

Demonstration of Use Cases Implementable with the Super-large AI EXAONE... Collaboration Plans Announced

At an event held via YouTube, Baek Kyunghoon, head of LG AI Research Institute, shared use cases that can be realized with LG’s super-large AI 'EXAONE', including the recently unveiled AI artist 'Tilda', as well as collaboration plans with partner companies.

Super-large AI is an AI capable of thinking, learning, and making judgments like humans by self-learning large-scale data based on computing infrastructure capable of large-scale computation.

'Tilda' is the first AI human implemented with EXAONE, who collaborated with designer Park Yoonhee to showcase clothing at New York Fashion Week on the 14th. The 200 garments revealed that day were produced based on over 3,000 images and patterns created by Tilda.

Baek explained, "Tilda is a representative case demonstrating the possibility of AI and humans collaborating in fields requiring creativity."

Baek also demonstrated an advanced chatbot model that understands emotions embedded in customers’ language, converses naturally like a human, and summarizes content at the level of a professional counselor, as well as deep document understanding (DDU) technology being researched with LG Chem and Elsevier.

Deep document understanding technology is based on multimodal AI that handles both language and vision, analyzing not only text such as papers and patents but also visual elements to quickly and accurately analyze and accumulate data on new drugs and new materials.

The first AI human 'Tilda' implemented by LG AI Research Institute (Photo by LG AI Research Institute)

LG Accelerates to Secure Global Super-large AI Ecosystem

LG AI Research Institute also revealed plans to provide various services that allow partner companies to easily and conveniently utilize super-large AI.

First, within the first half of this year, the institute plans to provide partners with application programming interfaces (APIs) that enable easy development of customized expert AI, removing barriers to entry for development. The APIs can be used in eight areas: Q&A, text classification, summarization, conversation, text generation, keyword extraction, image generation, and image captioning.

Additionally, to enable those without coding expertise or AI developers to easily use EXAONE on the web, the institute plans to launch the service platform 'EXAONE Playground' in the second half of the year, accelerating the popularization of super-large AI.

LG AI Research Institute also unveiled 'EXAONE-Tuning', a new technology that reduces the burden of data security and AI development time for partner companies.

Because super-large AI has an enormous number of parameters in its artificial neural network during training, additional learning requires significant time and resources. Accordingly, most companies researching super-large AI are developing technologies that use fewer parameters while maintaining performance.

EXAONE-Tuning, developed by LG AI Research Institute, reduces the number of parameters used by up to 90% compared to the most commonly used methods, making it much lighter and faster while increasing accuracy nearly fourfold.

For example, by additionally training EXAONE only on financial sector language, specialized expert AI services like AI bankers can be created.

This is possible because EXAONE has already learned the world’s largest corpus of 600 billion units, over 250 million high-resolution images combined with text, and has learned professional data to a certain level through collaboration with LG affiliates.

Baek Kyunghoon, head of LG AI Research Institute, said, “It is meaningful that we have established a platform for creating a global super-large AI ecosystem where EXAONE and partner companies grow together and innovate business models. Through close cooperation with partners, we will create top 1% expert AI in all industrial fields, provide completely new customer experiences never seen before, and become a leader driving the popularization of super-large AI.”

This content was produced with the assistance of AI translation services.