Summary of Key Points from the Conference Call Industry Overview - The focus is on the AI edge technology that empowers mobile devices, enabling local AI model operations while ensuring data privacy and low-latency interactions. This technology is primarily applied in AI smartphones and AI PCs, requiring robust local computing power and multimodal content processing capabilities [1][2][3]. Core Insights and Arguments - **AI Smartphone Projections**: It is anticipated that global shipments of AI smartphones will reach 54% by 2028, with the Chinese market expected to hit 150 million units by 2027. Initially, AI features will be introduced in the high-end market before gradually penetrating the mid-to-low-end segments [1][5]. - **AI PC Development**: AI PCs are positioned as core platforms that integrate computing power, personal large models, and applications while protecting user data privacy. The penetration rate of AI PCs is expected to rise, driving an increase in average selling prices (ASP) by 10%-15% for PCs with integrated NPU [1][6]. - **Smart Automotive Sector**: The demand for NPU and customized storage solutions is significantly increasing in smart vehicles, particularly for offline model deployment in cabin domains. Independent NPUs are becoming standard to ensure large model inference and interaction in areas without signal coverage [1][7]. - **NPU as a Focus for Chip Manufacturers**: NPU, designed specifically for AI, is a key focus for chip manufacturers. It excels in scalar, vector, and tensor computations, maximizing user experience in generative AI applications through heterogeneous computing alongside CPU and GPU [1][8]. Additional Important Content - **NPU Performance Metrics**: The main performance indicators for NPU include TOPS (Tera Operations Per Second) and memory bandwidth, which are crucial for inference response capabilities. NPU is widely used in smartphones, PCs, and automotive applications [1][9]. - **Current and Future NPU Forms**: Currently, NPU is primarily integrated within processors or SoCs. However, discrete NPU solutions are being explored to enhance computing power and optimize battery life during standby [1][10]. - **Market Potential for Discrete NPU**: If all flagship smartphones adopt discrete NPU solutions, the expected shipment volume could reach around 100 million units for third-party high-end smartphones. In the PC market, a 10% penetration rate for high-end products would require approximately 20 million discrete NPUs, leading to a potential total shipment of 120 million units across smartphones and PCs [1][11]. - **Customized Storage Solutions**: Customized storage is critical for achieving optimal NPU performance, similar to HBM for GPUs. The market for customized storage is projected to reach $2-3 billion in the next two to three years, especially as discrete NPU solutions penetrate the PC and smartphone markets [1][12][14]. - **Competitive Landscape**: In the NPU and customized storage sectors, domestic companies like Xiaomi, Honor, and Lenovo are actively developing NPU solutions, while international players include Qualcomm and MediaTek. In customized storage, companies like Gigadevice are leading in performance, with traditional DRAM manufacturers also participating [1][15].

图片来源： Minecraft 随着传统的人工智能基准测试技术显得力不从心， AI 构建者正转向更具创意的方法来评估生成式 AI 模型的能力。 对一群开发者而言，这个新舞台便是微软旗下的沙盒建造游戏—— Minecraft 。 网站 Minecraft Benchmark （或 MC-Bench ）是合作开发的，旨在让 AI 模型在直接挑战中相互竞争，以响应提示并创建 Minecraft 作品。 用户可以投票决 定哪个模型做得更好，只有在投票后才能看到每个 Minecraft 作品是由哪个 AI 制作的。 对于发起 MC-Bench 的 12 年级学生 Adi Singh 来说， Minecraft 的价值并不在于游戏本身，而在于人们对它的熟悉程度。毕竟，它是有史以来最畅销的视 频游戏。 即使是没有玩过游戏的人，仍然可以评估哪个方块化的菠萝表现得更出色。 " Minecraft 让人们更容易看到 AI 发展的进展，" Singh 告诉 TechCrunch 。"人们已经习惯了 Minecraft ，习惯了它的外观和氛围。" MC-Bench 目前列出了八名志愿者贡献者。根据 MC-Bench 的网站， Ant ...