Published on August 05, 2026 at 06:01 CEST (UTC+2)
libexpat now funded by the City of Munich for up to 6 months (176 points by spyc)
The blog post announces that the maintainer of libexpat, a widely used C XML parser, is now funded for up to six months by the City of Munich through its Open Source Sabbatical program. For the first time, maintaining libexpat is his regular paid job rather than a side activity competing with work and personal life. His top priorities include fixing five known unfixed vulnerabilities, adding support for XML 1.0r5, and improving the project's robustness and maintainability. He notes that work has already begun on a vulnerability reported by Mozilla, and he is employed remotely by digitial@M. The post also hints at ongoing challenges with unvalidated AI-generated submissions.
Pi's Minimalism Is Its Advantage (170 points by luispa)
Pi is a minimalist coding harness that deliberately ships with only four tools and a system prompt/tool definitions under 1,000 tokens, in contrast to increasingly complex AI coding platforms. The author argues that AI has made code cheap, but larger prompts and layers of orchestration make tools expensive and not necessarily better. Evidence from users and case studies with Databricks and Shopify suggests vanilla Pi produces strong results. Databricks' own benchmark of coding agents on a multi-million-line codebase highlights how performance should be evaluated against real-world tasks and cost. The post frames minimalism as a feature that improves both performance and affordability.
Eight Myths on Software Engineering and GenAI (110 points by tchalla)
This ACM Queue article examines eight myths surrounding generative AI in software engineering. It challenges common assumptions about what AI coding tools can and cannot do, including overblown claims about productivity gains and code reliability. The piece advocates for a more evidence-driven view of GenAI, emphasizing human oversight and realistic evaluation. It serves as a counterpoint to both uncritical adoption and outright dismissal.
After Losses, Retail Investors Flock to 3x Leverage as 2x Product Are Restricted (26 points by mapping365)
The Asiae article reports that retail investors are moving into 3x leveraged products after 2x leveraged products were restricted. Despite suffering losses, these investors appear to be seeking even higher risk/return exposure, likely in South Korea's retail trading market. The shift raises concerns about investor protection and the risks of leveraged products during volatile markets. This is more of a financial news story, but it illustrates how retail behavior can amplify risk after regulatory changes.
Mistral's Shieldstral: 3B open-weights model for multimodal moderation (335 points by riadsila)
Mistral released Shieldstral, a 3B-parameter open-weights model built for multimodal content moderation. Unlike traditional guardrail models with fixed harm taxonomies, it accepts plain-language policy descriptions at inference time and treats moderation as a question-answering task. It reportedly matches or outperforms models up to 7x larger on text safety and sets a new state of the art on multimodal moderation, while running on a single 16GB GPU. The model is released under Apache 2.0 and supports both text and image safety evaluation without retraining. This makes customizable safety classifiers more accessible and efficient.
IP and DNS Leaks in WebKit Affecting Proxy Browsers and iCloud Private Relay (59 points by lapcat)
Security researchers from Mysk found that three WebKit features—DNS prefetching, WebAuthn related-origin requests, and WebTransport—can bypass proxy configuration in WebKit-based browsers on iOS and macOS. That means a webpage can cause the device to make direct network connections, revealing the user's real IP address and DNS traffic. This affects privacy-focused proxy browsers, including Tor browsers on iOS and their own Psylo, as well as Apple's iCloud Private Relay. The researchers fixed the issues in Psylo 1.3.1 and built a proof-of-concept detection site. The article highlights the difficulty of ensuring network-level privacy on top of a shared browser engine.
Zigbee vs. Matter over Thread:Understanding IoT Protocol Performance in Practice (40 points by teleforce)
This arXiv paper presents an experimental comparison of Zigbee and Matter over Thread under real-world smart-home conditions. Zigbee is described as a mature, non-IP mesh networking solution, while Matter over Thread is an IP-based architecture aimed at interoperability across ecosystems. The authors note that despite extensive design documentation, there is little empirical performance data comparing the two. They conduct tests on a testbed to measure protocol behavior in practice. The work aims to give IoT developers practical performance insights beyond theoretical specifications.
DuckDB – Data power tools for your laptop, now in Clojure (2023) (65 points by sourdecor)
TechAscent's post explains why they turned to DuckDB for large data processing in Clojure. Their existing in-memory columnar platform, tech.ml.dataset, becomes unwieldy for datasets like ~100GB of CSV files, while JDBC/Postgres row-to-column conversion is inefficient. DuckDB offers an out-of-memory, columnar, local relational engine that fits their functional programming model. They describe integrating DuckDB with Clojure and with formats like parquet and arrow to handle larger-than-memory data. This reflects a broader trend of using embedded analytical databases instead of distributed cluster frameworks.
Show HN: Simple algorithm and color space to generate diverse skin tones (478 points by automatoney)
This Show HN project defines a color space specifically designed to generate diverse skin tones for inclusive color tools. It includes a JavaScript color
In Memory of My Wife, Elise Cawley, with Thanks for 36 Wonderful Years (1083 points by jdcampolargo)
Analysis generated by deepseek-reasoner