"NVIDIA to Support Korean AI Ecosystem"...Sharing Nemotron Expertise

Keynote Speech by NVIDIA Vice President Bryan Catanzaro
Free Distribution of 7 Million Persona Dataset
First Korean Demonstration of 'Build a Claw'

Global artificial intelligence (AI) company NVIDIA has announced a dataset specifically tailored for the Korean market, demonstrating its strong commitment to supporting the domestic AI ecosystem.

Bryan Catanzaro, Vice President of Applied Deep Learning Research at NVIDIA, introduced the 'Nemotron Personas Korea' dataset during his keynote speech at Front1 in Mapo, Seoul, on April 21. Catanzaro explained, "This dataset was built based on data from Statistics Korea and unique linguistic and cultural statistics of Korea." He emphasized that "it consists of 7 million fully synthetic personas designed to reflect the demographic distribution of the real world, while ensuring that no personally identifiable information is included."

NVIDIA plans to distribute this dataset under a permissive license. Catanzaro stressed, "We hope this dataset will be a practical resource for Korean AI developers to build more useful models optimized for Korean users."

On this day, NVIDIA also shared real-world technical know-how accumulated from developing its open large language models (LLMs), Nemotron 3 'Super' and 'Ultra.' Catanzaro highlighted the intelligence of the dataset, beyond just hardware performance, as a key know-how. He stated, "Through smart dataset composition, we were able to reduce language model pre-training time by about four times compared to before, even with the same hardware environment." He added, "This also creates efficiency by enabling the model to use fewer tokens to answer questions during inference." In addition, Catanzaro introduced NVIDIA's own optimization techniques that address both energy efficiency and performance, such as maintaining high accuracy during model pre-training.

Catanzaro also discussed the necessity of building 'specialized AI.' He explained, "AI is not a universal solution that applies equally to every organization," and added, "In order to solve the unique problems faced by each company and organization, it is essential to build specialized AI that safely integrates proprietary know-how and internal data specific to that organization." He continued, "We are very pleased to collaborate with Korea's dynamic AI research institutes such as Naver, SK Telecom, KT, LG, and Krafton," adding, "The experience and assets of the Nemotron project will serve as a powerful foundation for Korea's AI industry to move towards an open and accelerated future."

Additionally, NVIDIA introduced its autonomous AI agent-building program, 'Build a Claw,' to Korean developers for the first time. First unveiled at NVIDIA's annual developer conference 'GTC 2026' last month, this program was held as an 'execution-focused' session, where developers could directly install and run their own AI agents on actual hardware within an hour.

Based on the combination of the local-first AI agent 'OpenClaw' and the secure 'NVIDIA NemoClaw' stack, the agent—equipped with persistent memory and file access permissions—was able to remember users' past conversations and set its own work goals. In addition, the secure NVIDIA Nemotron model was applied, demonstrating the robust security of the local AI environment by maintaining data ownership while protecting personal information.

EPA Yonhap News

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