缔零科技完成数千万元融资，用AI替代人工内容审核，成本降60%

Core Insights - Dili Technology, focusing on cognitive security, has completed a multi-million yuan angel round financing to enhance its core product "Dili Rules" and build a cognitive security community [2] - The AI content security market in China is projected to reach 20 billion yuan in 2024, with an annual growth rate exceeding 35% [2] - Traditional manual review methods are becoming inadequate to handle the exponential growth of AI-generated content, prompting a shift towards proactive defense strategies in AI safety [2][4] Company Developments - Dili Technology's product "Dili Rules" achieved an 80% accuracy in model review by September 2024, with plans to reach 99.5% accuracy by March 2025 [2] - The company has established deep collaborations with major entities like People's Daily and Wuhan University, serving over 15 large enterprises including Douyin and Xiaohongshu [2] - The technology has demonstrated a threefold increase in efficiency for risk governance on platforms, reducing strategy adjustment time from three days to one [2] Product Features - The company employs a "dual-wheel drive" strategy, providing comprehensive review services for UGC platforms and acting as a "safety steward" for AIGC platforms [3] - "Dili Rules" can complete complex content reviews in 30 seconds, reducing costs by 60% compared to traditional manual reviews [3] - The technology supports the identification of 30 risk types with a response speed significantly faster than human reviewers [3] Strategic Vision - Dili Technology aims to create a sustainable cognitive security ecosystem, with a current focus on B2B clients, which are expected to maintain a 70:30 revenue split with B2C clients [5] - The company is exploring user needs in the C-end market and plans to introduce cognitive security products for individual users to protect vulnerable groups from harmful information [4][5] - The long-term vision includes a shift in the revenue structure, anticipating an increase in C-end contributions to 70% while reducing B-end to 30% [5]