NIA / Ministry of Science and ICT
National Public AI Training Data Initiative
Public AI training data is created for openness and sustained use, which makes ethical, safety, and representational standards just as important as data quality.
dda contributed to the development of Korean and multilingual datasets for public-purpose AI, supporting more reliable use across Korea’s AI research and industry ecosystem.
- Designing data structures and standards suitable for public-purpose AI training
- Establishing quality management frameworks that support diverse downstream uses
- Applying language processing standards that account for privacy and ethical considerations
- Defining operational processes that enable validation and auditability
- Designing scalable structures to support long-term and sustained use
dda contributed to the development of Korean and multilingual datasets for public-purpose AI, supporting more reliable use across Korea’s AI research and industry ecosystem.
- Designing data structures and standards suitable for public-purpose AI training
- Establishing quality management frameworks that support diverse downstream uses
- Applying language processing standards that account for privacy and ethical considerations
- Defining operational processes that enable validation and auditability
- Designing scalable structures to support long-term and sustained use
- AI Response Quality Evaluation Data (Ministry of Science and ICT/NIA, 2023)
- Human-AI text conversation + AI utterance evaluation labeling data for RLHF
- Human-AI text conversation + AI utterance evaluation labeling data for RLHF
- Korean BlenderBot Data (Ministry of Science and ICT/NIA, 2022)
- Topic-based daily conversation data from approximately 1,400 designed personas
- Topic-based daily conversation data from approximately 1,400 designed personas
- Multilingual Colloquial Translation Corpus Data (Ministry of Science and ICT/NIA, 2021)
- Korean ↔ French/Spanish/German and other multilingual translation parallel data construction