Published on January 19, 2026 at 18:01 CET (UTC+1)
Kiel Institute Analysis: US Americans pay 96% of tariff burden (361 points by 47282847)
A Kiel Institute analysis of 2025 US tariffs concludes that American importers and consumers bear 96% of the cost, with foreign exporters absorbing only about 4%. The research, based on shipment-level data, shows near-complete pass-through of tariffs to US import prices, causing a $200 billion surge in customs revenue paid by Americans. When faced with tariffs, foreign exporters maintained their prices and reduced shipment volumes instead of lowering them.
GLM-4.7-Flash (141 points by scrlk)
This article announces GLM-4.7-Flash, a new 30B parameter sparse Mixture-of-Experts (MoE) AI model from Z.ai, positioned as a top performer in its class. It provides benchmark scores showing it competes with or exceeds similar models like Qwen3-30B and details instructions for local deployment using popular inference frameworks like vLLM and SGLang. The model is presented as an efficient option for lightweight deployment.
CSS Web Components for marketing sites (18 points by zigzag312)
The author argues that standard web components with Shadow DOM are unsuitable for marketing site design systems due to their mandatory JavaScript dependency. They propose "HTML Web Components" as an alternative architecture that uses plain HTML enhanced with custom elements for interactivity, aligning with goals of progressive enhancement and performance for users with low-powered devices or poor connectivity.
The Microstructure of Wealth Transfer in Prediction Markets (15 points by jonbecker)
This research analyzes over 72 million trades on the Kalshi prediction market, finding systematic inefficiency contrary to the efficient market hypothesis. It identifies a wealth transfer mechanism where impulsive "Takers" overpay for affirmative outcomes, paying an "Optimism Tax," while patient "Makers" profit by selling into this biased flow. The effect is most pronounced in high-engagement categories like Sports and Entertainment.
Show HN: Pipenet – A Modern Alternative to Localtunnel (6 points by punkpeye)
Pipenet is a modern, open-source tunneling tool presented as an alternative to Localtunnel for exposing local servers to the internet. It supports HTTP, WebSocket, SSE, and streaming protocols, offers features like custom subdomains and TypeScript/ES Modules, and can be used via a public service, self-hosted, or integrated via an SDK. It is designed for contemporary deployment environments.
A decentralized peer-to-peer messaging application that operates over Bluetooth (428 points by no_creativity_)
BitChat is a decentralized, peer-to-peer messaging application that operates over Bluetooth mesh networks, requiring no internet, servers, or phone numbers. Each device acts as a node, automatically discovering peers and relaying messages across multiple hops to create an ad-hoc network. It is designed for censorship resistance, surveillance resistance, and functionality during internet outages, with clients available for iOS, macOS, and Android.
Folding NASA Experience into an Origamist's Toolkit (39 points by andsoitis)
This NASA Spinoff profile details how former JPL engineer Robert Lang applied mathematical tools from his work on lasers and optoelectronics to the field of computational origami. The algorithms he developed for folding complex 3D paper structures have since found significant commercial and scientific applications, including in medical stents and space telescope mirrors, demonstrating cross-disciplinary innovation.
Radboud University selects Fairphone as standard smartphone for employees (407 points by ardentsword)
Radboud University has selected Fairphone as its standard corporate smartphone, citing sustainability, cost efficiency, and manageability. The Fairphone's design prioritizes easily replaceable parts and fair/recycled materials to extend lifespan. While new issues will primarily be Fairphones, the university will also reissue returned Samsung devices where possible, and existing iPhones can be used until end-of-life.
Iterative image reconstruction using random cubic bézier strokes (13 points by luthenwald)
This is a showcase and repository for "splined," a Rust-based project that performs iterative image reconstruction using randomly generated cubic Bézier strokes. The process is accelerated using Apple's Metal graphics API. Users can generate stylized, non-photorealistic interpretations of input images, with results varying based on random seeds.
Ask HN: COBOL devs, how are AI coding affecting your work? (102 points by zkid18)
This Hacker News "Ask HN" thread solicits experiences from COBOL developers on how AI coding tools are affecting their work. The consensus among commenters is that AI (particularly current LLMs) has limited impact due to compliance restrictions, inability to run locally on legacy systems, poor handling of COBOL's strict formatting, and, crucially, a lack of understanding of deep, undocumented business logic that defines these critical systems.
Trend: The Rise of Efficient, Deployable Model Architectures. The launch of GLM-4.7-Flash, a high-performing 30B MoE model, highlights a focus on creating models that balance capability with practical deployment costs (inference speed, memory). This matters because the real-world utility of AI depends on affordable and scalable deployment, not just benchmark scores. The implication is a competitive push toward specialized, efficient model families that serve specific use-cases or resource constraints.
Trend: AI as a Creative and Analytical Co-Pilot, Not an Autopilot. The "splined" image reconstruction tool and the COBOL developer discussion represent two sides of the same coin: AI excels at generating novel patterns (artistic strokes) or automating simple, tedious tasks (generating test data), but struggles with deep, context-specific logic and precise execution. This matters because it defines the current assistive role of AI—augmenting human creativity and handling grunt work, while leaving complex system understanding and final integration to humans.
Trend: Growing Tension Between AI Capability and Legacy System Realities. The COBOL thread starkly illustrates that the vast "installed base" of critical software is largely insulated from current AI coding aids due to technical debt, compliance, and unique business logic. This matters because it creates a bifurcated market: rapid AI-driven development for greenfield projects versus slow, human-centric maintenance for legacy systems. The takeaway is that significant economic value lies in creating AI tools specifically tailored to understand and navigate legacy code environments.
Trend: Data Reveals Behavioral Biases, Not Just Rational Markets. The prediction market analysis uses large-scale trade data to expose systematic human irrationality (like an "Optimism Tax"), challenging the notion that crowdsourced prices are purely efficient. This matters for AI/ML as it underscores the importance of modeling real human behavior and bias in financial or forecasting AIs. It suggests opportunities for AI systems that can identify and arbitrage these behavioral inefficiencies, or for designing better market mechanisms.
Trend: Edge Computing and Decentralization as an AI/ML Adjacent Frontier. The popularity of BitChat (Bluetooth mesh networking) and the discussion of performance for low-powered devices in the web components article signal a strong interest in robust, offline-first, and decentralized systems. This matters for AI as it pushes development toward smaller, more efficient models that can run on edge devices (like phones) and operate in disconnected or low-bandwidth environments, which is crucial for global accessibility and resilience.
Trend: Cross-Disciplinary "Toolkit" Transfer Accelerates Innovation. The story of NASA's math for lasers being repurposed for computational origami, which later aided engineering design, is a paradigm for AI progress. Foundational algorithms and mathematical insights often have unforeseen applications. This matters because it encourages investment in fundamental AI research and creating flexible "toolkits." The implication is that breakthroughs in one domain (e.g., a new neural architecture) can unpredictably catalyze advances in seemingly unrelated fields.
Trend: Sustainability is Becoming a Tangible Tech Procurement Driver. Radboud University's institutional switch to Fairphone is a concrete example of ethical and environmental sustainability influencing technology decisions at an organizational level. This matters for AI/ML due to the enormous energy and resource costs associated with large-scale model training and deployment. The trend will increase pressure to develop more energy-efficient hardware and algorithms, and to consider the full lifecycle environmental impact of AI infrastructure.
Analysis generated by deepseek-reasoner