Articles

This article provides an in-depth analysis of the Sora App launched by OpenAI. Despite reaching one million downloads within 5 days, the app's user retention rate plummeted to 1% within 30 days, accompanied by a significant drop in its App Store ranking. Drawing on data from a16z partner Olivia Moore, industry observations, and user feedback, the author reveals the primary drivers behind Sora App's user attrition. Key issues include suboptimal recommendation algorithms, the absence of built-in sharing functionalities, restrictions on uploading real avatars, and the intrusive impact of watermarks. These shortcomings have hindered its transformation from a mere AI tool into a product with genuine social attributes. The article also references insights from OpenAI Chief Research Officer Mark Chen, who suggests that OpenAI, as a dedicated AI research company, faces inherent difficulties in product operation, ultimately predicting that Sora App will likely persist as a creative tool but gradually fade as a social product.

The article reports in detail on the joint announcement by XPENG Motors and Huawei at the Songshan Lake Park in Dongguan. The XPENG X9 super extended-range version will be the first to feature Huawei's DriveONE fully integrated high-voltage oil-cooled generator. This collaboration aims to address common issues in current extended-range models, including insufficient pure electric range, power degradation in low-battery conditions, and NVH (Noise, Vibration, and Harshness) concerns, particularly for large MPVs. Through joint development, they achieved an industry-leading power density of 1.88kW/kg and a generation efficiency of over 92%. They also introduced innovative technologies such as multi-jet gear oil cooling, resolverless algorithm, and active cylinder deactivation, effectively controlling the noise of the range extender to within 0.5dB(A), barely perceptible to the human ear. The system supports 800V high-voltage fast charging, achieving a combined range of 1602 kilometers. The article also delves into Huawei's strategic transformation from a traditional component supplier to a technology co-creator, emphasizing how its experience in participating in vehicle development informs its component solutions, making them more valuable at the system level.

The article provides a detailed review of the grand occasion of the '2025 Jiazhi Gravity Annual Summit'. The conference attracted 2000+ attendees, and 70+ technology industry guests shared in-depth insights on AI development. Zhang Yijia, the founder of Jiazhi Guangnian, released the '30 Judgments on the 2025 AI Industry' report, proposing that AI has entered the 'Default AI' stage, characterized by non-selectivity, low perception, and system binding, and provided a detailed interpretation from four dimensions: technology, product, industry, and society. In terms of technology, the emphasis is on economic viability and the qualitative change of the AI memory mechanism; products are moving towards 'headless' and 'invisible agents'; on the industry side, Business-to-Business (B2B) competes for 'Automation Hub' while Business-to-Consumer (B2C) competes for 'Super Gateway', and Embodied AI welcomes a turning point in application. The social level focuses on the polarization of employment and the AI curse. In addition, the conference also featured two summit dialogues, where experts engaged in vigorous discussions on the academic origin of AI, computing power optimization, multi-modal reasoning, the development path of Embodied AI, and industrial implementation strategies, discussing the optimization path of Chinese AI under limited computing power, the application prospects of reinforcement learning, breakthroughs in multi-modal and World Model, and the differences in B2B and B2C implementation. Parallel sessions focused on AI Agents, Embodied AI, AI infrastructure and ecosystem construction, technology industry investment, etc., with multiple company founders sharing Agent technology, physical AI paradigm shifts, the construction of Chinese AI supply chains, Embodied AI evaluation standards, the challenge of building strong relationships with users for B2C AI Agents, the explainability of AI reasoning, and the new logic and experience of technology investment. Finally, the article also mentions that Jiazhi Guangnian released the three major lists of [Jiazhi 100], [Jiazhi Cool Vendor], and [2025 China Technology Industry Investment List], aiming to recognize the pioneering enterprises and individuals that promote the development of China's technology industry.

The article reports that independent cloud provider Vultr announced an investment of over $1 billion to build a 50-megawatt (MW) AI chip cluster equipped with 24,000 AMD Instinct MI355X AI accelerators in Ohio, scheduled to be operational in the first quarter of 2026. This move aims to provide customers with more cost-effective AI computing power, with service prices typically half those of large data center operators. Vultr secures funding from AMD and multiple banks by purchasing chips and operating in leased data centers. Despite concerns about industry bubbles and GPU depreciation, Vultr's CEO is confident that AI computing infrastructure is still severely insufficient and believes GPUs have a reasonable lifespan.

This article highlights Google Cloud's initiative to make PubMed's vast biomedical literature available as a BigQuery public dataset, enhanced with Vertex AI's vector search capabilities. It addresses the limitations of traditional keyword searches in medical research, which often miss critical connections, and the time-consuming manual literature reviews. The solution facilitates semantic search, moving beyond simple keyword matching to identify relevant medical concepts. Key use cases demonstrated include transforming oncological literature reviews at Princess Máxima Center for Pediatric Oncology, enabling rapid clinical decision support, accelerating drug discovery and repurposing, improving public health services, aiding rare disease diagnosis, and enhancing insurance and payer intelligence. The article provides a practical SQL example for performing semantic search on PubMed Central articles within BigQuery and points to further resources for agentic literature review. It positions this integration as a democratizing force for medical knowledge, aligning with NIH's mission.

This article provides an in-depth analysis of the diverse definitions of a 'good project.' The author starts by emphasizing that 'good' is subjective. It depends on individual priorities, such as freedom, enjoyment, or financial gain. Next, the article suggests that if solely oriented towards 'money,' projects that seem accessible but have low barriers to entry and allow average people to quickly profit are actually 'good projects,' using CPS (Content Subscription Service) and Jingdong Coupon Codes as examples. The author refutes the view that 'low-barrier projects have low ceilings,' emphasizing that through building teams and SOP-driven operations, even accessible projects can achieve scalable profitability. The article compares the advantages and disadvantages of 'high-end' projects versus 'profitable for average people' projects, and explains that the author's own choice of the AI Product track is due to the pursuit of 'freedom' and 'fun.' Finally, the article guides readers to combine their own values with objective profitability standards to find 'good projects for you,' and includes a promotion for the 'Shengcai Youshu (a Chinese business community)' community.

The article deeply analyzes how Vipshop, a vertical e-commerce platform, has achieved profitability for over fifty consecutive quarters in the fierce market competition by focusing on the branded discount sales model, becoming a 'hidden champion' in the e-commerce sector. The article elaborates on Vipshop's early key transformation from luxury discounts to mid-to-high-end fashion brand sales, as well as its constructed moat, including a professional buying team, JIT/JITX supply chain management, and a high-value SVIP user system. However, facing drastic changes in the external environment, such as price wars from comprehensive e-commerce giants, the rise of content e-commerce, refined inventory management by brands, and the simplicity of its own business structure, Vipshop's revenue and active user numbers have declined, and the growth boundary of its 'small but beautiful' model has become increasingly clear. The article concludes by discussing Vipshop's attempts to expand beyond apparel and accelerate its global presence. Its future success will test whether a vertically focused model can overcome its limitations.

Taking the recent viral 'cane-stealing' activity as an example, this article deeply analyzes the success secrets of such 'self-service' gamified business models. By reviewing the popularity of 'stealing vegetables' games like Happy Farm and Ant Forest, the author points out that users' desire for 'controlled transgression' has never ceased. The article explains how the 'cane-stealing' activity provides a 'thrill of feeling like a thief' experience within legal and moral boundaries, satisfying people's psychological need to transcend daily life and release stress. This model not only helped farmers clear their sugarcane inventory but also brought unique entertainment and emotional value to visitors. The article emphasizes that creating 'rebellious pleasure' within safe boundaries is a hidden code of the contemporary emotional economy, and argues that this innovative experience represents people's rebellion against fixed lifestyles and their pursuit of free choice. Additionally, the article also touches upon potential management issues when such models attract excessive crowds, such as visitors eating produce on-site without paying.

The article analyzes the global hardware market fluctuations revealed by tech analyst Moore's Law Is Dead. The core point is that OpenAI secretly signed two contracts with Samsung and SK Hynix on October 1, 2025, securing 40% of the global DRAM supply. This unexpected strategic move, in the absence of market warning and a lack of safety stock, quickly triggered panic buying by other AI giants, OEMs, and cloud service providers, leading to an epic surge in the prices of memory (DRAM) and solid-state drives (SSDs). DDR5 memory prices surged by 156% in just three weeks. The lack of safety stock was due to several factors: tariff uncertainties, continuously falling DRAM prices discouraging manufacturers from stocking up, and geopolitical issues preventing the circulation of second-hand DRAM manufacturing equipment. The author notes that OpenAI is purchasing raw, uncut, and unpackaged DRAM wafers of unspecified standards. This suggests their aim is not immediate use, but rather to create artificial scarcity by seizing the supply chain, weakening competitors and maintaining their lead in AI. The article concludes by providing purchasing advice for affected hardware in the next 6-9 months and calling for more attention to OpenAI's market behavior, sparking intense debate about the AI bubble and market disruption.

This article focuses on the year 2025, providing an in-depth analysis of Shanghai's rapid ascent in the field of Embodied AI. This is primarily due to its optimized ecosystem driven by a Service-oriented Government. By opening over a hundred core scenarios, including high-end manufacturing, medical and healthcare, and urban governance, Shanghai offers up to 40 million yuan annually in Computing Power Coupons and 5 million yuan in Data Corpus construction support. Furthermore, the formation of an industrial cluster in Zhangjiang Robot Valley ensures close proximity between upstream and downstream partners. The article highlights that these initiatives have spurred three major technological advancements in Embodied AI: Cognitive Awakening, Tactile Reconstruction, and Matrix Evolution. Finally, the article highlights the GDPS 2025 Global Developer Pioneers Summit as a comprehensive practical demonstration of Shanghai's Embodied AI ecosystem achievements, encompassing industry, service, and rescue tracks, emphasizing practical capabilities and technological maturity.