NC launches LLM performance verification model 'VARCO Judge LLM'

Paper on LLM Evaluation Model Published in Global Conference... Proving Global Technology Level

On the 23rd, NCSoft announced the launch of 'VARCO Judge LLM,' the first evaluation model in Korea to verify the performance and execution capabilities of large language models (LLM).

VARCO Judge LLM is an evaluation model that examines how quickly and accurately other LLMs perform tasks. Recently, companies have been spending a lot of time searching for the most suitable model among various types and sizes of LLMs available in the market.

By using this model, companies can more effectively verify the LLM model best suited for their AI services. In other words, it allows them to validate the performance level of their own LLM against the evaluation model to prove superiority over other models or quickly identify weaknesses and reinforce them.

VARCO Judge LLM boasts excellent capabilities in solving LLM bias problems and outstanding Korean language performance. It has the highest performance among models of its class. Notably, NCSoft also demonstrated its technological prowess on the global stage by publishing a paper on the LLM evaluation model at EMNLP (Empirical Methods in Natural Language Processing), a world-renowned natural language processing (NLP) conference, this year.

As the first domestic game company to develop its own language model, NCSoft actively utilizes AI technology throughout the game development process and overall internal work efficiency. Through the release of this evaluation model, the company plans to improve the quality of its proprietary LLM, VARCO, and secure a leading position in the field of LLM model evaluation.

Lee Yeon-su, head of NCSoft's Research Division, said, "In the rapidly evolving AI market, services that select and apply the optimal model for each industry are becoming increasingly important. VARCO Judge LLM will establish itself as an essential tool in AI business, going beyond merely improving the quality of existing LLM-based services."