ENGLISH
    • 한국어
    • ENGLISH
    • 日本語
    • 中文-繁體

    2024.09.23 Press Release

    NCSOFT Launches ‘VARCO Judge LLM’, Korea’s First Evaluation Model for LLM Performance

    NCSOFT introduces ‘VARCO Judge LLM’, designed to assess the performance and capabilities of large language models (LLMs).

    Research on ‘LLM evaluation model’ was presented at the prestigious global conference EMNLP, underscoring NCSOFT’s technological leadership.

    “Our goal is to improve service quality and development efficiency in AI business through the LLM evaluation model.”

    NCSOFT today announced the launch of ‘VARCO Judge LLM’, Korea’s first evaluation model designed to assess the performance and capabilities of large language models (LLM).

     

    VARCO Judge LLM is an evaluation model that measures how quickly and accurately various LLMs can perform tasks. With the growing variety of LLMs available in the market, companies often face challenges and spend considerable time in selecting the most suitable models for their needs. By utilizing the VARCO Judge LLM, it allows businesses to more effectively verify which LLM models are best aligned with their AI services.

     

    NCSOFT’s VARCO Judge LLM boasts superior capabilities in addressing LLM bias issues and demonstrates exceptional performance in the Korean language, ranking it highest among equivalent models. In recognition of its technological leadership, NCSOFT presented a paper on the LLM evaluation model at one of the most prestigious NLP conferences, EMNLP (Empirical Methods in Natural Language Processing), further showcasing its expertise on the global stage.

     

    By leveraging NCSOFT’s evaluation model, AI-based service providers can quickly compare and assess various LLMs, ensuring they adopt the most optimized models for their services. AI model developers can use the VARCO Judge LLM to verify their own models’ performance, demonstrate superiority over competing models, or identify and address areas for improvement. Additionally, model hub businesses can streamline the process of selecting, optimizing, and deploying LLMs to offer services more effectively.

     

    As the first Korean game company to develop its own language model, NCSOFT actively integrates AI technology throughout the game development process and the overall internal operational efficiency. With the release of this evaluation model, NCSOFT aims to further improve the quality of its proprietary LLM, ‘VARCO’, while solidifying its leadership in the LM evaluation field.

     

    Yeonsoo Lee, Head of NC Research, stated, “In the rapidly evolving AI market, selecting and applying the optimal model for each industry is becoming increasingly important. VARCO Judge LLM will not only improve the quality of existing LLM-based services but will also become an essential tool for the broader AI business landscape.”

     

    For more information, please visit the official NCSOFT website.