Latest news with #ExaonePath2.0
Korea Herald
2 days ago
- Business
- Korea Herald
LG aims for global AI elite with next-gen Exaone platform
Following founder's Cabinet appointment, LG AI Research's new leaders set sights on global top-three status LG AI Research unveiled on Tuesday its evolving Exaone ecosystem, introducing a series of its latest and newly developed Exaone-based AI models and services co-developed with global partners. At the 'LG AI Talk Concert 2025' held at LG Sciencepark in Seoul, the institute showcased its lineup of AI innovations built on proprietary foundation models, including Exaone 4.0, Korea's first hybrid AI model combining reasoning and generative capabilities and Exaone Path 2.0, a precision medical AI system. To secure global competitiveness, the institute aims to expand its Exaone across all major industries, including biotechnology and finance. "The evolution of agentic AI directly leads to physical AI, a technology that transforms the real world. LG AI Research will accelerate the development of physical AI technology, enabling AI to perceive, judge, and alter actual environments,' Lee Hong-lak, co-president who was newly appointed Monday, said. While LG largely kept its Exaone models out of the public's reach as a B2B product, the company first unveiled its cutting-edge Exaone 4.0 model via API through the Friendly AI inference platform for companies and institutions, enabling real-world business integration. Another major unveiling was Exaone 4.0 VL, a multimodal vision-language model. It is capable of interpreting complex professional documents, images and molecular structures. LG AI Research said the model outperformed Meta's Llma 4 Scout in internal benchmarks, positioning it as a key component of the Exaone ecosystem – serving as its 'eyes' in real world applications. Also unveiled for the first time include the beta version of ChatExaone, an enterprise AI agent designed for secure internal use and previously limited to LG Group employees. The beta is currently available only to enterprise or business accounts. Exaone Data Foundry, a platform that streamlines the production of high-quality training data, and Exaone On-Premise, a full-stack enterprise AI solution built entirely with Korean hardware and software for deployment in secure and independent environments, were also unveiled for the first time during the event. Arman Sahovic, APAC Head of Data Platform Solutions at the London Stock Exchange Group, introduced business intelligence services developed with Exaone. Sahovic explained that LSEG and LG AI Research are collaborating to create a service that predicts investment asset return trends and generates reports based on LSEG's data, news and disclosure information, thereby aiding investment decisions globally. At the event, chief researchers at LG AI Research expressed their ambition to become one of the global top three players in the AI field. Their immediate goal is to lead in generative and agentic AI, with a longer-term vision to pioneer the era of physical AI, where systems can reason autonomously and take action to solve real-world problems. 'The race to secure AI technology is no longer just a corporate issue — it has expanded into a competition between nations. Without a self-sustaining ecosystem, dependency deepens, and we risk falling behind. Building our own ecosystem is essential,' said Kim Yoo-chul, head of Strategic Division at the institute. Lee Hwa-young, vice president leading AI Biz Dev and DX Group, also urged for government support in infrastructure. 'To build good AI, you need data, talent, technology and infrastructure. LG has everything except infrastructure,' Lee said. 'The government is now stepping in to build that infrastructure, and we are working hard to stay at the forefront of technology and become one of the global top three in AI.' The institute's two co-Presidents succeeding Bae Kyung-hoon following his appointment as Minister of Science and ICT, also made their first official appearance since their appointments. Lim Woo-hyung, one of the co-presidents, said he was proud of Bae's new role and vowed to continue leading the institute with dedication. 'Our goal is not just to advance AI technology, but to apply it directly in industrial settings and create real business value. We're actively exploring how to strengthen AI capabilities across LG affiliates and accelerate digital transformation,' Lim said. Co-head Lee Hong-lak said, 'I will work to elevate LG AI Research to global excellence. We will continue developing Exaone into a world-class technology, support its adoption across the group and help uncover new growth engines.'
Korea Herald
09-07-2025
- Health
- Korea Herald
LG's new AI slashes genetic testing time to under 1 minute
LG AI Research on Wednesday unveiled its next-generation artificial intelligence model, Exaone Path 2.0, which can diagnose diseases by analyzing patients' pathology images alone. This new AI model is expected to drastically reduce the time required for genetic testing from two weeks to under one minute, marking a step forward in the early diagnosis of diseases such as cancer. Exaone Path 2.0 has been upgraded from the first edition, trained on higher-quality data. It offers enhanced capabilities in analyzing and predicting genetic mutations, gene expression patterns and subtle structural features within human tissues and cells, the AI research center, affiliated with tech giant LG Group, said. 'Using Exaone Path 2.0, we can reduce the time needed for genetic testing from over two weeks to under one minute,' said Park Yong-min, a lead researcher at LG AI Research. 'Doctors and pharmaceutical companies can quickly analyze pathology images of a patient's tumor tissue, identify which genes have mutated and match them to suitable targeted therapies in real time.' The latest model has been trained on multiomics data — including DNA and RNA — paired with pathology images, allowing it to understand both cellular structures and underlying biological mechanisms of disease, LG AI Research explained. As Whole Slide Images often span several gigabytes, AI models typically break them into smaller patches to reduce processing loads. This, however, can lead to 'feature collapse,' where broader structural context is lost. Exaone Path 2.0 addresses this by learning from both detailed patches and the full-slide view. Trained on over 10,000 WSI-multiomics pairs, it achieves a state-of-the-art 78.4 percent accuracy in predicting genetic mutations, LG explained. LG also introduced specialized models for diseases such as lung and colorectal cancer, which are designed to help clinicians identify eligible patient groups for targeted therapies while avoiding unnecessary tests. With the goal of pioneering clinical AI implementation, LG AI Research has partnered with Professor Hwang Tae-hyun and his team at Vanderbilt University Medical Center, one of the top biomedical research institutions in the US. Together, they aim to develop a multimodal medical AI platform to redefine precision medicine, LG said. Unlike conventional approaches that first develop technologies and then seek clinical application, the research institute is working on a clinic-first strategy, focusing on solving real-world problems directly within clinical settings for AI development. 'Our goal is not simply to develop a new AI model, but to build an AI platform that can be actively used by medical professionals in real clinical settings to support diagnosis and treatment,' Hwang said. Hwang is a Korean-American scientist who leads the stomach cancer initiative under the US government's Cancer Moonshot program. He also founded the Molecular AI Initiative at Vanderbilt University Medical Center, where he conducts interdisciplinary research at the intersection of AI and molecular medicine. With Hwang's team, LG AI Research said it plans to start with oncology and later expand its multimodal AI research into areas such as transplant rejection, immunology and diabetes. The research institute has also teamed up with the Jackson Laboratory in the US to discover biomarkers and develop treatments for Alzheimer's disease, and it is working with Professor Baek Min-kyung's team at Seoul National University to develop next-generation protein structure prediction AI.
