PeopleofSuperintelligence

Researcher

Kang Min Yoo

Also known as 유강민, Yoo Kang-min

Kang Min Yoo is a natural language processing researcher known for his work on NAVER's HyperCLOVA and HyperCLOVA X language models. After a computer science PhD on data augmentation, he joined NAVER's AI Lab in July 2020 and co-authored the 2021 study introducing HyperCLOVA, an 82-billion-parameter Korean variant of GPT-3. His GPT3Mix method, also from 2021, used large language models to generate realistic training text, and in January 2023 he pushed NAVER to form a task force on instruction following. He was the first listed author of the HyperCLOVA X Technical Report in April 2024. By June 2025 he led Foundation Research at NAVER Cloud, where he oversaw the evaluation of the reasoning model HyperCLOVA X THINK.

Now at
Naver

Timeline

  1. 2020

    A PhD on data augmentation

    He finishes a computer science PhD in machine learning and natural language processing, with a dissertation on data augmentation, and joins NAVER's AI Lab in July 2020, where he also works with the Conversation team on chatbot technology. At EMNLP 2020, he and five co-authors, including Sang-goo Lee, present the Variational Hierarchical Dialog Autoencoder, which generates goal-oriented dialogs to augment training data for dialog state tracking.

  2. 2020–2021

    HyperCLOVA, a Korean GPT-3

    His team starts work on HyperCLOVA, NAVER's hyperscale AI model, in September 2020. He is among the co-authors of the EMNLP 2021 study that introduces HyperCLOVA, an 82-billion-parameter Korean variant of GPT-3 trained on a Korean-centric corpus of 560 billion tokens.

  3. 2021

    GPT3Mix for text augmentation

    With Dongju Park, Jaewook Kang, Sang-Woo Lee and Woomyoung Park, he publishes GPT3Mix in Findings of EMNLP 2021. The method uses large language models to generate realistic text samples from a mixture of real samples and adds soft labels predicted by the model.

  4. 2022

    Ground-truth labels in prompts

    At EMNLP 2022, he is first author of a study that re-examines how correct labels in demonstrations shape in-context learning. The paper introduces two metrics, Label-Correctness Sensitivity and the Ground-truth Label Effect Ratio.

  5. January 2023

    An instruction-following task force

    As ChatGPT draws Korean users, he urges his team in January 2023 to prepare NAVER's own model with instruction following. He persuades the organization to form a task force in January 2023.

  6. April 2024

    The HyperCLOVA X technical report

    In April 2024, he is the first listed author, with 395 co-authors, of the HyperCLOVA X Technical Report, which describes a family of large language models tailored to the Korean language and culture. As the NAVER Cloud leader responsible for the report, he presents its multilingual reasoning and translation tests as evidence for the potential of sovereign AI.

  7. 2025

    The KMMLU benchmark

    In 2025, he co-authors KMMLU in the NAACL 2025 proceedings, a Korean benchmark of 35,030 expert-level multiple-choice questions across 45 subjects, collected from original Korean exams.

  8. June 2025

    Leading Foundation Research

    In June 2025, NAVER unveils its reasoning model HyperCLOVA X THINK, and he is in charge of its performance evaluation as leader of Foundation Research at NAVER Cloud. He reports notable results in visual reasoning tasks and plans to advance the model toward stronger multimodal reasoning.

  1. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes Naver. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  2. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  3. The United States employs an estimated 48% of people shaping AIWe estimate where the people shaping AI work today, how universities, companies and research institutes split across countries, and how country shares have changed since the 1960s.
  4. 43% of living people shaping AI work outside their country of birthWe estimate that 43% of living people shaping AI work outside their country of birth (520 of 1,198 people with known birth and employment countries).

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Kang Min Yoo yet. Send us a free-licensed photo.

  1. Technology that opens a new world navercorp.com. Checked 20 October 2026.
  2. Variational Hierarchical Dialog Autoencoder for Dialog State Tracking Data Augmentation aclanthology.org. Checked 20 October 2026.
  3. What Changes Can Large-scale Language Models Bring? Intensive Study on HyperCLOVA: Billions-scale Korean Generative Pretrained Transformers aclanthology.org. Checked 20 October 2026.
  4. GPT3Mix: Leveraging Large-scale Language Models for Text Augmentation aclanthology.org. Checked 20 October 2026.
  5. Ground-Truth Labels Matter: A Deeper Look into Input-Label Demonstrations aclanthology.org. Checked 20 October 2026.
  6. [2404.01954] HyperCLOVA X Technical Report arxiv.org. Checked 20 October 2026.
  7. HyperCLOVA X Technical Report Reaffirms its Superiority Over Global Models, Open Source AI, and Translators navercorp.com. Checked 20 October 2026.
  8. KMMLU: Measuring Massive Multitask Language Understanding in Korean aclanthology.org. Checked 20 October 2026.
  9. NAVER Unveils “HyperCLOVA X THINK,” a Reasoning Model with Superior Linguistic Capabilities navercorp.com. Checked 20 October 2026.
  10. Naver unveils homegrown AI model HyperClova X Think koreajoongangdaily.com. Checked 20 October 2026.