CrowdWorks Wins Government Contract to Build Korean AI Reasoning Dataset

Дата публикации: 08-07-2026 12:58:36

CrowdWorks has won a government contract to build AI training data under the Ministry of Science and ICT's 2026 AI Hub Training Data (Inference) Project, administered by the National Information Society Agency (NIA). The company said on July 7 that it will lead the "Complex Document-Based Knowledge

Основное содержимое страницы с новостью.

CrowdWorks has won a government contract to build AI training data under the Ministry of Science and ICT's 2026 AI Hub Training Data (Inference) Project, administered by the National Information Society Agency (NIA). The company said on July 7 that it will lead the "Complex Document-Based Knowledge Inference Data" project in a consortium with Wert Intelligence and Zendive.

The project aims to develop high-quality AI training data to improve Korean-language reasoning capabilities in domestic and global generative AI models.

CrowdWorks said evaluations of Korean-language expertise among domestic and overseas AI models found that most still perform below the average human level, citing limitations in contextual understanding and the ability to derive complex causal relationships. The company said those shortcomings have become a barrier to AI adoption in real-world industrial applications.

To address the issue, the government has launched 10 projects to build advanced reasoning datasets. The CrowdWorks-led consortium will be responsible for developing the "Complex Document-Based Knowledge Inference Data" dataset.

According to CrowdWorks, the consortium will collect and refine multimodal public documents combining text with visual elements such as tables, graphs and diagrams, producing more than 10,000 high-quality reasoning training data samples. The project will establish a Korean-language multimodal complex reasoning dataset.

The company said it will analyze reasoning weaknesses across fields including public administration, science and technology, and laws and patents, then design sophisticated reasoning chains and high-difficulty questions using techniques such as Chain of Thought (CoT), Tree of Thoughts (ToT) and Graph of Thoughts (GoT). The goal is to train AI models on human-like reasoning processes rather than simply providing answers.

To reduce AI hallucinations, CrowdWorks plans to map verifiable sources to every step of the reasoning process and apply a hybrid validation pipeline combining automated large language model (LLM) verification with expert review to ensure data integrity.

Схожие новости

#	Наименование новости	Тональность	Информативность	Дата публикации
1	Samsung SDS, Samsung C&T Pursue New AI Data Center in Chuncheon	0	5	08-07-2026
2	Naver Cloud, Mistral AI Partner to Target Manufacturing AI Market	0	5	08-07-2026
3	Neosem Wins 8.6 Billion Won Memory Tester Order From Samsung Electronics	5	7	09-07-2026
4	Laserssel Wins 17.9 Billion Won Semiconductor Equipment Order	5	7	08-07-2026
5	Bad Taxes Would Slow AI Innovation	0	7	19-06-2026
6	What China’s HBM Catch-Up Should Teach Korea	0	5	02-07-2026
7	РФПИ и правительство намерены вместе развивать сферу искусственного интеллекта	0	0	05-09-2019
8	Govt, Nasscom working on AI curriculum for undergraduate programmes	0	7	09-07-2026
9	Минцифры России расширит свои полномочия в сфере ИИ	0	5	04-07-2026
10	Ученые получат на развитие инфраструктуры суперкомпьютерных центров в СНГ 30 млн рублей	0	0	01-08-2019

Классификация: Пресс-релизы. Схожих патентов: 0. Схожих новостей: 10. Тональность: 0. Информативность: 5. Источник: thelec.net.