Naver AI Tab: 2x Faster, 30% Less Hallucination

Naver has integrated a new lightweight model into its AI Tab, significantly enhancing its AI search service. This new model offers twice the speed and reduces the hallucination rate by 30 percentage points compared to the previous HyperCLOVA X model. Naver announced these updates during a recent Tech Deep Talk session. The company highlighted key technologies for next-generation AI search, including this optimized model for conversational AI, along with harness engineering and multimodal technology. Director Ki-chang Lee emphasized that this model is specifically optimized for Naver services, from data building to reinforcement learning. The model is a product-native large language model, incorporating Naver’s data, service scenarios, and user feedback. It focuses on data, architecture, and training to maximize efficiency. This means faster response times and improved accuracy for users.