Naver Advances AI Search, Cutting Hallucinations and Doubling Speed

*Editor's Note: 'AI PRISM' (Personalized Report & Insight Summarizing Media) is an "AI-based personalized news recommendation and summary service" developed with support from the Korea Press Foundation. It selects and provides six customized news items by reader type.

1. Naver AI Search Cuts Hallucinations by 30 Percentage Points and Reduces Costs to One-Third

- Key Summary: Naver has applied three next-generation technologies to its conversational AI search "AI tab." The "product-native LLM," a lightweight model optimized for search services based on Naver's own model HyperCLOVA X, changed its structure so that computational load increases in proportion to question length, curbing response-time increases even when processing long documents. It also applied "clarity reinforcement learning," which asks back to confirm intent on ambiguous questions, succeeding in reducing hallucinations by 30 percentage points compared to HyperCLOVA X. With a "division-of-labor SLM" structure that combines role-specialized SLMs (small language models), equipment operating costs were lowered to as low as one-third of previous levels, and response speed was improved by more than twofold.

2. 'Video AI Powerhouse' China's Kling Raises 4.67 Trillion Won, Facing Off Against Seedance

- Key Summary: Kuaishou, the parent company of Chinese video AI firm Kling, raised up to $3 billion (approximately 4.67 trillion won) through a third-party allocation paid-in capital increase, with Kling valued at about $18 billion (approximately 27.54 trillion won). This investment is a Pre-IPO in nature, as Kuaishou spins off Kling to begin Hong Kong stock market listing procedures within 12 months, and China's three major big tech firms—Baidu, Alibaba, and Tencent—all participated. Kling has surpassed 100 million users worldwide, and about 70% of its revenue comes from overseas subscriptions and API (application programming interface) calls for 50,000 corporate clients. First-quarter revenue this year was 650 million yuan (approximately 146.5 billion won), a surge of more than 300% year-on-year.

3. Fraudulent Startup Subsidy Claims to Be Caught by AI Starting This Year

- Key Summary: The Ministry of SMEs and Startups and the Korea SMEs and Startups Agency will introduce an AI-based anomaly detection system within the year to block fraudulent claims of Youth Startup Academy startup subsidies. The Youth Startup Academy provides promising founders with an average of 70 million won and up to 100 million won in commercialization funding, with more than 100 billion won in government budget invested annually. The system has AI learn from approximately 300,000 project cost data records and about 900,000 attached documents to automatically detect related-party transactions, unit-price outliers, duplicate claims, and abnormal labor-cost data. When an abnormal transaction is detected, a risk score and detailed information are displayed on a dashboard and an alert is provided to the person in charge, with the final judgment made by that person.

4. Mirae Asset Life Insurance (085620), 3 of Every 4 New Contracts Undergo 'Pre-Screening'

- Key Summary: Mirae Asset Life Insurance was found to have an average pre-screening rate of 75.5% among new contracts over the recent three months (March to May). Pre-screening is a digital underwriting (insurance enrollment screening) system that determines in advance, based on a customer's medical history data, whether insurance enrollment is possible. Since its introduction in 2023, applications in the first quarter of this year reached 478,444, more than double the figure a year earlier. Last month, Mirae Asset Life became the first in the industry to introduce an LLM (large language model)-based generative AI underwriting system, "AI-FIT." AI analyzes a customer's health information and insurance claim history to present screening results in real time, and interprets diagnoses to automatically extract the KCD (Korean Standard Classification of Diseases).

5. LG Chem (051910), First Supply of Semiconductor Stripper to U.S. Amkor

- Key Summary: LG Chem said Thursday it will begin its first mass-production supply of semiconductor strippers to Amkor, a global leader in semiconductor back-end processing in the U.S. A stripper is a core process material that removes photoresist (PR) and residues left on substrates after semiconductor circuit formation. LG Chem entered the semiconductor market based on technological capabilities accumulated in display strippers. The supplied product is customized and optimized for Amkor's new line, shortening photoresist and residue removal time by 50% from previous levels. Meanwhile, LG Chem announced a strategy in March to more than double its electronic materials business and is expanding its semiconductor packaging materials portfolio, including copper clad laminate (CCL), die attach film (DAF), and photosensitive insulating material (PID).

6. From Semiconductor Inspection Equipment to AIDC... "Targeting 20 Billion Won in Combined Revenue and Orders"

- Key Summary: TeraView, a KOSDAQ-listed British company, is expanding the application scope of its inspection equipment from semiconductor packaging inspection to the AI data center (AIDC) market. Its flagship product "EOTPR" uses terahertz technology to capture even micro-defects difficult to detect with X-ray, with Intel, Nvidia, Samsung, and AMD among its major clients. TeraView recently won a 2.6 billion won project from a global AI chip company to develop large-area panel inspection equipment inside AI data center racks. The company aims to achieve more than 20 billion won—double the current level combining 6.3 billion won in cumulative revenue and 5.1 billion won in order backlog—by the end of the April settlement next year.