Successful adoption of AI requires automation of data generation and processing, efficient infrastructure expansion, and the development of customized models for each industry.

At the Artificial Intelligence Semiconductor Future Technology Conference held at the Lotte Hotel in Sogong-dong on December 10, Jeon Ki-jeong, head of LG AI Research, gave a presentation on the topic of 'AX's success: EXAONE AI Stock Strategy Case Study.'

Vice President Jeon Ki-jeong diagnosed, “Along with the performance improvements of large-scale language models such as GPT, the scaling laws of data, models, and infrastructure are continuously applied, and we have now entered the era of ‘agent AI,’ where AI goes beyond simple content creation and is responsible for complex decision-making.”

Citing a recent McKinsey report, Jeon Ki-jeong, the head of the electric vehicle division, pointed out that many companies are still at the proof-of-concept (PoC) stage when it comes to AI adoption.

“To maximize the effectiveness of AI, strategic investments are needed at each stage, including data collection, infrastructure expansion, and application development,” emphasized Jeon Ki-jeong, head of the electric vehicle division.

In particular, he revealed that securing high-quality data and parallelizing and optimizing infrastructure such as GPUs are the keys to success.

'EXAONE', developed by LG AI Research Center, is a large-scale Korean-centric language model that has been recognized for its high performance in global benchmarks.

EXAONE 4.0, which performed pre-training on data worth 14 trillion tokens, is evaluated as one of the 'global top 3' both domestically and internationally and is being applied in various industrial fields.

“During the development of EXAONE, there were numerous trials and errors and technical challenges, including data quality management, architecture innovation, and GPU parallelization,” said the former head of the division.

Additionally, “Within the LG Group, the chatbot ‘Chat EXAONE’ is being used for various tasks such as document-based QA, web document analysis, coding, and data analysis,” he said. “With three times more users than expected, we have maximized the efficiency of infrastructure operation by optimizing GPU utilization to over 95%.”

In manufacturing, AI agents have been introduced to naphtha plants to automate complex decision-making, such as blending raw materials and controlling furnace temperatures.

In actual fields, productivity innovation is being achieved by adhering to schedules provided by AI almost 100%.

In LG Display's quality improvement case, the AI agent 'Heidi' was used to analyze the cause of defects and suggest measures, resulting in a reduction in the quality improvement period from three weeks to two days.

In the financial sector, we collaborate with the London Stock Exchange to list only large-cap stocks. We also developed a system that automatically generates AI-powered prediction and analysis reports for small and medium-sized stocks. This achieved higher returns compared to existing prediction models.

He also introduced that data learning, combining expert knowledge and AI automation, is rapidly taking place in complex service areas such as the National Pension Service.

In the future, it is expected that faster and more efficient AI services will spread through collaboration with semiconductor technologies such as NPU and cloud and application developers.