TLDR: Generative AI startup 42Maru has made history by becoming the first company in Korea to receive the ‘AI-MASTER’ reliability certification for its large language model, LLM42. This achievement positions 42Maru as the sole entity in the nation to hold both public and private AI reliability certifications.
Seoul, South Korea – 42Maru, a leading generative artificial intelligence (AI) startup, announced today that its proprietary large language model, LLM42, has been awarded the prestigious ‘AI-MASTER’ reliability certification. This significant milestone marks LLM42 as the first model in the Republic of Korea to achieve this private sector certification, further solidifying 42Maru’s position at the forefront of AI innovation and trustworthiness.
This latest certification builds upon 42Maru’s previous success in November 2024, when it became the first large language model in Korea to pass the ‘AI Reliability Certification (CAT)’ by the Telecommunications Technology Association (TTA), a public certification system operated by the Ministry of Science and ICT. With the acquisition of the ‘AI-MASTER’ certification, 42Maru now stands as the only company in Korea to possess both public and private AI reliability certifications, setting a new benchmark for AI development in the country.
The ‘AI-MASTER’ system is Korea’s inaugural private AI reliability certification, a collaborative effort developed by the Korea Artificial Intelligence Industry Association (AIIA), Suresoft Technologies, and SelectStar. This robust certification framework is meticulously designed, drawing upon the European Union’s seven principles of ‘Trustworthy AI’ and nine ISO/IEC international standards. The evaluation process involves quantitative assessments across 63 distinct items, encompassing document reviews, rigorous testing, and data-based analyses. Notably, over 60% of these evaluation items are assessed using automated tools, ensuring a high degree of objectivity and consistency in the certification process.
During the comprehensive evaluation, 42Maru’s LLM42 garnered particularly high scores in critical areas such as personal information protection, effective bias response, and overall technical stability. LLM42 is distinguished as a lightweight, ultra-large language model specifically optimized for the Korean language environment. Its core characteristics include a low-cost operational structure, enhanced security features, and broad applicability across various industrial sectors.
One of LLM42’s key innovations lies in its ability to significantly mitigate the ‘hallucination’ problem commonly associated with generative AI. This is achieved through the integration of its proprietary retrieval-augmented generation (RAG42) technology and AI reading comprehension (MRC42) technology. Furthermore, LLM42 supports a private mode, enabling the secure processing of sensitive data, which is particularly beneficial for corporate clients handling confidential information.
Also Read:
- NongHyup Bank Unveils Advanced Generative AI Platform to Revolutionize Financial Operations
- KAIST Unveils NPU Technology Dramatically Boosting Generative AI Performance by Over 60%
Kim Dong-hwan, CEO of 42Maru, emphasized the importance of this achievement, stating, “In the midst of the widespread adoption of generative AI technology, reliability and responsibility are the key criteria for sustainable growth. LLM42’s official certification is a case that simultaneously proves the safety and reliability of the technology, and we will continue to focus on providing stable AI infrastructure to the public and industry in general.” This dual certification underscores 42Maru’s commitment to developing secure, reliable, and responsible AI solutions for the evolving digital landscape.
