Seoul National University held a gap ceremony for the ‘Super-Gigantic AI Model and Platform Optimization Center’ and started constructing a man-made intelligence (AI) model in earnest.

On the twentieth, Seoul National University held a gap ceremony for the ‘Super-Giant AI Model and Platform Optimization Center (CHAMP)’ following the choice of the Engineering Field Leading Research Center (ERC) project by the Ministry of Science and ICT on the auditorium of the Cultural Center. The event attracted great attention with a lecture by Professor Andrew Ng of Stanford University.

Following the choice of the ERC project, Seoul National University CHAMP will promote execution platform, algorithm, and application development research to beat the constraints of super-large AI model technology from this yr to February 2030. A complete of 14.05 billion won is invested in the middle, including 13.5 billion won from the federal government, 400 million won from Seoul National University, 100 million won from the day after tomorrow, and 50 million won from green onions.

CHAMP goals to construct its own model and develop technology by lowering its dependence on large language models (LLMs) and technologies concentrated in overseas big tech. It plans to develop and disclose an AI model that is simple to learn and infer, and an execution platform for it that’s small in size and similar in performance to a super-large AI model, and demonstrated. Finally, the goal is to spread the research results to the industry.

To this end, Lee Jae-jin, an adjunct professor on the Department of Computer Science and Engineering at Seoul National University and the pinnacle of the CHAMP Center, announced step-by-step goals. Aiming to develop a light-weight AI model based on Korean and English and a software and hardware platform for its learning, ▲Step 1 (2023-2026) builds a model with similar accuracy to GPT-3.5 and one-hundredth the scale. ▲Phase 2 (2027-2029) develops a model with similar accuracy to the most recent AI model and one-hundredth the scale.

Professors from the Graduate School of Data Science at Seoul National University will lead the middle, and researchers with business experience will take part in conducting research on parallel processing, model weight reduction, and AI application, that are essential for LLM development.

Lee Jae-jin, head of the middle, said, “Overall optimization was difficult because of the shortage of communication between subjects researching AI model applications, algorithms, and platforms in Korea.”

Reporter Juyoung Lee juyoung09@aitimes.com