en.Wedoany.com Reported - South Korean portal site Daum launched its 'AI Summary' service on the 1st of this month, officially entering the AI search competition dominated by Naver and Google. All three services utilize their respective large language models (LLMs), competing on the core pillars of high-quality answers, ease of use, and accuracy.
Daum's 'AI Summary' beta service analyzes web documents directly through its LLM to provide users with comprehensive search results. Previously, Naver and Google had already launched 'AI Briefing' and 'AI Overviews' respectively, displaying summarized answers alongside links to original content.
Naver's AI Briefing is based on HyperCLOVA X, Google's AI Overviews on Gemini, and Daum uses Upstage's 'Solar' model. Possessing proprietary LLMs allows these three services to better control workflows and enhance application flexibility. Among them, Solar is continuously optimized through the government project 'Independent AI Foundation Model,' in which Upstage participates as a supervisor; HyperCLOVA X focuses on the Korean language, having learned from a vast amount of Korean data; Gemini leverages multimodal technology capabilities and extensive resources within the Google ecosystem.
Naver emphasizes that the advantage of its AI Briefing lies in providing refined summary answers based on data accumulated over nearly 30 years of operating lifestyle services such as blogs, cafes, and Knowledge iN, as well as the latest information within its ecosystem. Daum positions 'ease of use' as the core selling point of its AI Summary. A representative from its operating company, AXZ, stated that the introduction of the Solar model enables more accurate and faster searches, with future plans focusing on practical, vertical specialized fields. Google, meanwhile, emphasizes the accuracy and reliability of AI Overviews, which is designed to display only information that can be supported by top web results, avoiding the generation of false information or fabricated facts.
Additionally, Google has improved its detection mechanism for contradictory queries, avoids including sarcastic or humorous content in answers, and has taken measures against overviews that violate content policies.
