CrowdWorks Leads AI Verification Market

**CrowdWworks and TTA Propose Standard for Generative AI Reliability Evaluation Framework** CrowdWworks has successfully completed the Korea Telecommunications Technology Association (TTA)'s 'Research Project on Practical Approaches to Generative AI Reliability Evaluation', significantly enhancing its expertise in the AI reliability field. The core of this project was to develop and demonstrate a standard framework for systematically evaluating the reliability and safety of generative AI. CrowdWworks led the practical demonstration of LLM (Large Language Model) reliability evaluation and the development of educational materials, conducting in-depth evaluations of three LLM models developed by domestic companies. Based on datasets, they meticulously analyzed response patterns for each model, identified potential risk factors, and designed various attack scenarios to intensively explore the models' vulnerabilities. The evaluation involved both automated AI model assessment and in-depth evaluation by a specialized LLM red team, selected from CrowdWworks' pool of 600,000 data experts. This red team applied comprehensive AI risk assessment criteria, including violence, illegality, irrationality, non-factual content, misleading information, and unethical behavior, to quantitatively and qualitatively analyze the risk of model responses and derive practical improvement measures. Based on the unique AI reliability evaluation expertise secured through this research, CrowdWworks plans to further advance its reliability evaluation services that minimize AI risks for businesses. Furthermore, this year, it plans to expand its AI service reliability evaluation business to various industries and establish strong leadership in the field of AI reliability and safety. Kim Woo-seung, CEO of CrowdWworks, emphasized, "The AI reliability evaluation framework developed through this TTA research project holds great significance as it presented an important standard for domestic generative AI reliability evaluation." He added, "Based on our network of 600,000 data experts and a verified evaluation system, we will actively support the development of safe and reliable AI services and lead the market."