LG AI Research seeks to address industrial challenges with 'expert' AI models

LG AI Research unveiled expert AI models based on EXAONE at LG AI Talk Concert 2026, saying the goal is to solve practical industrial problems in manufacturing, science and finance. The institute said the models are already producing tangible results in real-world settings. It also announced plans for an AI-powered autonomous laboratory.

EXAONE Tabular analyzes raw numerical data in tables and can predict future values with relatively small amounts of data.

LG AI Research said Tabular can cut the time needed to respond to model changes by 85 percent in manufacturing environments.

EXAONE Discovery screened 420,000 candidate substances for LG Household & Health Care and found ramsidil in one day.

LG AI Research said the usual discovery process for that kind of substance can take 22 months.

The institute also introduced EXAONE Business Intelligence for the financial sector and plans to build an AI-powered autonomous laboratory.

Published Sep 14, 2026 3:36 pm KST

Updated Sep 14, 2026 4:26 pm KST

Research head expresses doubts on Big Tech's call for AI development slowdown

LG AI Research co-head Lim Woo-hyung speaks during LG AI Talk Concert 2026 at LG Science Park in Gangseo District, Seoul, Monday. Courtesy of LG AI Research

LG AI Research, an artificial intelligence (AI) institute of LG Group, has unveiled “expert” AI models based on its EXAONE large language model, which are already delivering tangible results by addressing practical challenges encountered in real world industrial environments.

The models were unveiled during LG AI Talk Concert 2026, held by LG AI Research to introduce use cases of its expert AI models, each specialized in manufacturing, science and finance.

“Building a good AI model is important, but LG AI Research’s mission is to solve difficult problems that industries have struggled with for years,” LG AI Research co-head Lim Woo-hyung said during his keynote speech.

“There are already many general-purpose AI models in the world that generate excellent answers. But solving problems in industrial settings requires AI that can understand numerous variables, even the 1 percent of exceptional cases. AI can prove its value in industry only when it can demonstrate tangible results.”

LG AI Research officials demonstrate scalp and hair diagnosis during LG AI Talk Concert 2026 at LG Science Park in Gangseo District, Seoul, Monday. Newsis

As part of that approach, LG AI Research unveiled EXAONE Tabular and EXAONE Omni-Inspect.

Tabular specializes in understanding raw numerical data presented in tables. It analyzes relationships between numbers and predicts future values. Because it can make predictions with relatively small amounts of data, the model can adapt quickly to new manufacturing environments, cutting the time needed to respond to model changes by 85 percent.

Omni-Inspect is designed to identify defects in components by analyzing images captured by cameras. Even when new products or processes are introduced and the images change, the AI agent can continue operating without requiring retraining.

EXAONE Discovery is a model specializing in science. Working with LG Household & Health Care, LG AI Research used the model to discover ramsidil, a new substance that could help prevent hair loss, in just one day, as the model screened 420,000 candidate substances and identified the most suitable one. This reduced the discovery process that typically takes 22 months to a single day.

The institute also introduced EXAONE Business Intelligence for the financial sector and unveiled plans to build an AI-powered autonomous laboratory. Under the autonomous lab, a foundation model for materials development will predict the outcomes of new material synthesis, while robotic equipment conducts experiments. The AI will then learn from the results and design the next experiments, creating a continuous cycle.

LG AI Research co-head Lee Hong-lak, right, talks to Artificial Analysis co-founder Geroge Cameron during LG AI Talk Concert 2026 at LG Science Park in Gangseo District, Seoul, Monday. Yonhap

During the conference, LG AI Research co-head Lee Hong-lak expressed his doubts over whether companies are heeding the call made by global big tech leaders. Anthropic CEO Dario Amodei on Saturday urged AI companies to reduce the pace with which they improve their most advanced models, with OpenAI CEO Sam Altman and SpaceXAI CEO Elon Musk backing the call.

“I haven’t seen any evidence that companies are really slowing down the pace of development,” Lee said. “If anything, they have accelerated their computing and research and development efforts, and I don’t expect that to slow.” Lee said.

“When it comes to releases and deployment, however, I think they are becoming more cautious rather than publishing something every week, given concerns over model risks and cybersecurity attacks. I understand that they are being more careful in internal testing and review before release.”

Artificial Analysis co-founder Geroge Cameron speaks during LG AI Talk Concert 2026 at LG Science Park in Gangseo District, Seoul, Monday. Korea Times photo by Nam Hyun-woo

During the conference, Artificial Analysis co-founder Geroge Cameron said that even in the age of AI agents, the performance of the underlying model remains the most important factor, but cost efficiency, speed and flexibility are also becoming increasingly important as companies move from experimenting with AI to using it at scale.

“Many companies would go bankrupt if they used Anthropic’s Claude Fable for everything,” he said, noting that the model costs $8.75 per task on Artificial Analysis’ intelligence index. “Many use cases don’t need to use the most expensive leading model in terms of intelligence achieved.”

He noted that companies are increasingly weighing not only intelligence, but also speed, cost and flexibility, with open-weight models also offering additional benefits for firms seeking to run or fine-tune models on their own infrastructure.

Cameron said Korean AI models, including LG AI Research’s EXAONE, still trail leading U.S. and Chinese models in overall intelligence, but that does not rule them out as competitive options when those trade-offs are considered.

He also stressed that Korean developers should continue pursuing stronger general-purpose models and not lose sight of the global race toward artificial general intelligence.

LG AI研究院在2026年LG AI Talk Concert大会上发布了基于EXAONE平台的专家级AI模型，并表示其目标是解决制造业、科学和金融领域的实际工业问题。该研究院称，这些模型已经在实际应用中取得了显著成果。此外，研究院还宣布了建设人工智能驱动的自主实验室的计划。

EXAONE Tabular 分析表格中的原始数值数据，并能利用相对较少的数据预测未来的数值。

LG AI Research 表示，Tabular 可以将制造环境中对模型变更的响应时间缩短 85%。

EXAONE Discovery 为 LG 生活健康公司筛选了 420,000 种候选物质，并在一天内发现了 ramsidil。

LG AI 研究院表示，此类物质的常规发现过程可能需要 22 个月。

该机构还为金融行业推出了 EXAONE 商业智能，并计划建立一个人工智能驱动的自主实验室。

发布于2026年9月14日下午3:36（韩国标准时间）

更新于2026年9月14日下午4:26（韩国标准时间）

研究负责人对大型科技公司呼吁放缓人工智能开发表示怀疑

LG AI研究院联席负责人林宇亨周一在首尔江西区LG科学园举行的LG AI Talk Concert 2026上发表讲话。图片由LG AI研究院提供。

LG集团旗下的人工智能（AI）研究所LG AI Research推出了基于其EXAONE大型语言模型的“专家”AI模型，这些模型已经通过解决现实世界工业环境中遇到的实际挑战，取得了切实成果。

这些模型是在 LG AI Research 举办的 LG AI Talk Concert 2026 上发布的，旨在介绍其专家 AI 模型的应用案例，每个模型都专注于制造业、科学和金融领域。

LG AI Research 联席负责人林宇亨在主题演讲中表示：“构建一个好的 AI 模型固然重要，但 LG AI Research 的使命是解决各行业多年来一直面临的难题。”

“目前世界上已经有很多通用人工智能模型能够生成优秀的答案。但是，解决工业环境中的问题需要人工智能能够理解众多变量，甚至包括那1%的特殊情况。只有当人工智能能够展示切实可见的成果时，它才能在工业领域证明自身的价值。”

周一，在首尔江西区LG科学园举行的LG AI Talk Concert 2026上，LG AI Research的官员们演示了头皮和头发诊断技术。（Newsis）

作为该方法的一部分，LG AI Research 推出了 EXAONE Tabular 和 EXAONE Omni-Inspect。

Tabular 专门用于理解表格中呈现的原始数值数据。它分析数字之间的关系并预测未来值。由于它能够利用相对较少的数据进行预测，因此该模型可以快速适应新的制造环境，从而将模型变更的响应时间缩短 85%。

Omni-Inspect旨在通过分析摄像头拍摄的图像来识别组件缺陷。即使引入新产品或新工艺，图像发生变化，人工智能代理也能继续运行，无需重新训练。

EXAONE Discovery 是一款专注于科学领域的模型。LG AI Research 与 LG 生活健康合作，利用该模型仅用一天时间就发现了 ramsidil——一种可能有助于预防脱发的新物质。该模型筛选了 42 万种候选物质，并最终确定了最合适的成分。这使得通常需要 22 个月才能完成的发现过程缩短至一天。

该研究所还推出了面向金融行业的EXAONE商业智能系统，并公布了建设人工智能驱动的自主实验室的计划。在该自主实验室中，材料开发的基础模型将预测新材料合成的结果，而机器人设备则负责进行实验。人工智能将从实验结果中学习并设计后续实验，从而形成一个持续循环。

周一，在首尔江西区LG科学园举行的LG AI Talk Concert 2026活动上，LG AI研究院联席负责人李洪乐（右）与人工智能分析公司联合创始人乔治·卡梅隆交谈。（韩联社）

在会议期间，LG人工智能研究院联席负责人李洪乐对各公司是否响应全球大型科技公司领袖的呼吁表示怀疑。Anthropic首席执行官达里奥·阿莫迪周六敦促人工智能公司放慢改进最先进模型的速度，OpenAI首席执行官萨姆·奥特曼和SpaceXAI首席执行官埃隆·马斯克也支持这一呼吁。

李说：“我没有看到任何证据表明企业真的放慢了研发步伐。恰恰相反，他们加快了计算和研发投入，而且我预计这种趋势不会放缓。”

“然而，就发布和部署而言，考虑到模型风险和网络安全攻击，我认为他们变得更加谨慎，不再每周都发布新内容。据我了解，他们在发布前会进行更仔细的内部测试和审查。”

周一，人工智能分析公司联合创始人乔治·卡梅隆在首尔江西区LG科学园举行的2026年LG人工智能演讲音乐会上发表讲话。韩国时报摄影师南贤宇摄

在会议期间，人工智能分析公司联合创始人乔治·卡梅伦表示，即使在人工智能代理时代，底层模型的性能仍然是最重要的因素，但随着公司从人工智能实验转向大规模使用，成本效益、速度和灵活性也变得越来越重要。

“如果所有公司都使用Anthropic公司的Claude Fable模型，很多公司都会破产，”他说道，并指出该模型在人工智能分析公司的智能指数中，每个任务的成本为8.75美元。“很多应用场景并不需要使用智能水平最高的、价格最昂贵的模型。”

他指出，企业越来越重视的不仅是智能，还有速度、成本和灵活性，而开放权重模型也为那些希望在自己的基础设施上运行或微调模型的公司提供了额外的好处。

卡梅伦表示，包括 LG AI Research 的 EXAONE 在内的韩国人工智能模型在整体智能方面仍然落后于美国和中国领先的模型，但考虑到这些权衡因素，这并不排除它们作为有竞争力的选择。

他还强调，韩国开发者应该继续追求更强大的通用模型，不要忽视全球在通用人工智能领域的竞争。