Hacker News
1LinkedIn is searching your browser extensions
- Microsoft's LinkedIn is conducting large-scale corporate espionage by secretly scanning users' computers for installed software without consent or disclosure. The hidden code collects personal data revealing religious beliefs, political opinions, and job search activity of identified individuals, while mapping competitor software usage across companies. LinkedIn deceives EU regulators by providing restricted APIs while expanding surveillance of third-party tools, and shares collected data with external firms including an American-Israeli cybersecurity company.
Misleading headline framing
Users debate whether the headline "LinkedIn scans your computer" is misleading since it only scans browser extensions within the browser sandbox, not the entire computer. Some argue this distinction matters for public understanding, while others contend extension scanning still constitutes computer scanning.LinkedIn's justification and anti-scraping measures
LinkedIn defends the practice as necessary to detect extensions that violate terms of service and scrape user data without consent. Discussion centers on whether anti-scraping measures justify privacy invasion and scanning for religious/political extensions.Browser security and Chrome vs Firefox
Technical discussion about why browsers allow extension detection, with Firefox users noting they're protected due to randomized extension IDs while Chrome-based browsers remain vulnerable to this scanning technique.
2Decisions that eroded trust in Azure – by a former Azure Core engineer
- A Microsoft Azure engineer describes joining the Azure Core team in 2023 and discovering an organization planning to port 173 Windows agents to a tiny Linux-based accelerator card with severe hardware limitations. The author argues this impossible technical plan reflects deep organizational dysfunction that threatens Azure's reliability for critical customers like OpenAI and government systems.
Technical debt and code quality crisis
Azure reached a state where code refactoring was deemed too risky, with bug fixes rejected for fear of breaking systems. Engineers describe needing comprehensive test coverage before any changes, but corporate incentives don't reward cleanup work or maintenance - only new feature launches matter.Azure's reliability and user experience problems
Users report Azure feels like a "janky mess" with broken UI, AI-written documentation that's constantly wrong, and services that randomly fail. Engineers working across multiple clouds say 80-90% of their infrastructure issues come from Azure despite using it for basic workloads.Corporate incentives vs. engineering quality
Microsoft succeeds through contracts and sales rather than software quality. Non-technical decision makers choose Azure due to bundled Office deals and existing vendor relationships, while engineers who understand the technical limitations aren't involved in purchasing decisions.
3Sweden goes back to basics, swapping screens for books in the classroom
Tech in education effectiveness
Former EdTech worker argues pre-university tech use was a mistake, comparing it to Dutch famine effects. Others debate whether targeted tech literacy teaching vs wholesale digital adoption is better, with concerns about attention spans and cognitive development from excessive screen time.Physical vs digital learning materials
Finnish parent with 10 kids reports schools returning to paper books over screens. Discussion covers handwriting's cognitive benefits, distraction issues with tablets, cost differences between physical/digital textbooks, and e-ink readers as potential middle ground.Tech executives' parenting choices
Users note Steve Jobs limited his kids' iPad use, Zuckerberg restricts social media access for his children. Debate over whether this reveals hypocrisy or just normal parenting limits, and what it suggests about these technologies' safety for children.
4Show HN: Apfel – The free AI already on your Mac
- This content only contains a GitHub repository link to "Arthur-Ficial/apfel" with no additional information, description, or context provided about what the repository contains or its purpose.
Local AI privacy and security
Strong consensus that running AI models locally provides better privacy than cloud services. Discussion covers risks of data leakage, need for open training methods, browser security vulnerabilities, and encrypted inference options.Apple model performance and limitations
Mixed views on Apple's foundation models - some praise performance in specific tasks, but concerns about outdated training data, small context windows, and limited conversational abilities compared to newer models.Technical implementation and tooling
Discussion of various tools for accessing local models, including security considerations for network exposure, CLI interfaces, and integration with existing workflows like Alfred and KDE runners.
5Show HN: I built a frontpage for personal blogs
- Blogosphere is a platform that aggregates recent posts from personal blogs across various categories to help keep the independent web alive amid the dominance of social media and AI. It offers two versions - a minimal HN-inspired static version and a non-minimal version, and welcomes submissions of new blogs for review and approval.
Return to webrings and curated lists
Discussion about how the internet is regressing back to hand-curated lists and webrings due to declining search quality and AI content proliferation. Users debate whether this is positive, exploring blogrolls, web of trust systems, and community-maintained alternatives to centralized curation.Technical issues with the site
Users report pagination problems in the minimal version (can't access pages beyond 12) and infinite scroll issues preventing footer access in the modern version. The creator acknowledges these bugs and promises fixes while planning to add search to the minimal version.Discovery of similar blog aggregators
Community members share numerous alternative blog discovery platforms including minifeed.net, indieblog.page, various KB clubs, blogroll.org, and others. Discussion includes personal curation methods and the challenge of maintaining long-term interest in these aggregators.
6Tailscale's new macOS home
- Tailscale's menu bar icon on MacBooks with notches could disappear behind the camera notch, making it inaccessible to users. As a temporary fix, they added a popup warning when the icon was hidden. The real solution is their new windowed macOS interface that runs alongside the menu bar app, providing better access to Tailscale features and avoiding the notch issue entirely.
MacOS notch hiding menu bar icons
Users report frustrated customers and refund requests when the MacBook notch hides menu bar icons, making apps appear broken. Apple has no overflow menu solution despite this being an obvious UX problem that Windows solved decades ago with system tray management.Terminal commands to reduce menu bar spacing
Users share defaults commands to reduce spacing between menu bar icons, allowing 2x more items to fit. This workaround highlights how "just works" macOS requires technical knowledge to solve basic usability issues.Tailscale VPN setup recommendations
Discussion of using Tailscale for home network access, streaming geo-restricted content, and family device management. Users praise its ease of setup for non-technical family members and automatic connections.
7Lemonade by AMD: a fast and open source local LLM server using GPU and NPU
- Lemonade is an open source AI platform that runs locally on PCs, offering fast performance on GPUs and NPUs. It provides chat, image generation, and speech capabilities through a lightweight 2MB service with one-minute installation. The platform is OpenAI API compatible, works across Windows, Linux, and macOS, and supports multiple models simultaneously while maintaining privacy through local execution.
Lemonade vs Ollama performance comparison
Users discuss speed differences between Lemonade and Ollama for local LLM inference. Limited testing shows Lemonade completing tasks faster (1:14 vs 1:44 for qwen3.59b). Both use llama.cpp but Lemonade has AMD-specific optimizations and ROCm builds.NPU capabilities and limitations
NPUs are described as low-power, slow processors meant for small models under 8B parameters. Best suited for always-on tasks like speech processing rather than main chatbot workloads. Limited utility on desktop systems but useful for power efficiency.AMD hardware experience with ROCm
Mixed experiences with AMD's ROCm drivers - some users report "rough" experiences while others have stable setups. Discussion covers hardware lockups when exceeding VRAM, Vulkan vs ROCm performance differences, and AMD vs NVIDIA comparisons for local inference.
8Cursor 3
- Cursor 3 introduces a unified workspace for building software with AI agents, featuring a new interface built from scratch that allows multiple agents to work across different repositories simultaneously. The update includes seamless handoff between local and cloud agents, parallel agent execution, improved diff views for code review, and integration of existing Cursor IDE features like file browsing and plugins.
Serial vs parallel agent workflows
Users debate single-task AI assistance vs multi-agent swarms. Many prefer working on one feature at a time with AI help rather than parallel agents, citing better code quality and mental context management.Cost comparison between tools
Significant discussion around pricing differences between Cursor ($2k-7k/month) vs Claude Code Max (~$200/month) for similar productivity, with many users switching due to cost concerns.Code-first vs chat-first interfaces
Users express concern that Cursor v3's agent-centric UI moves away from developer-driven coding toward chat interfaces, preferring to maintain control over code review and architecture decisions.
9Artemis computer running two instances of MS outlook; they can't figure out why
- A humorous post by niki grayson imagines astronauts contacting mission control because their spacecraft computer is inexplicably running two instances of Microsoft Outlook, prompting NASA to prepare for remote technical support to resolve the issue.
Software reliability and security concerns in space
Discussion focuses on the risks of running unsealed Windows software in space that can download updates, send telemetry, and create unpredictable network traffic. Users debate whether NASA should use predictable, controlled systems instead of consumer software like Outlook for mission-critical environments.Technical alternatives to Outlook for space communication
Users debate email client options for space missions, weighing Outlook's familiarity against lighter alternatives like Thunderbird, Claws-mail, or command-line tools. Discussion covers bandwidth efficiency, offline capabilities, and the practicality of email for document transfer in space.Remote troubleshooting challenges and space networking
Technical discussion about the difficulties of remotely accessing spacecraft computers with ~380-420ms latency, WiFi signal detection possibilities from Earth, and comparisons to ham radio communication with the ISS.
10Artemis II crew take “spectacular” image of Earth
Camera technical details
Discussion about the Nikon D5 camera and AF-S 14-24mm lens used to capture the image, with EXIF data analysis revealing minimal Lightroom processing, high ISO 51200 settings, and comparison to newer Z9 model that was also brought on the mission.Earth nighttime photography
Conversation about how the image shows Earth's nightside illuminated by moonlight, creating a daylight-like appearance that confuses viewers, with visible aurora, city lights, and demonstrations of moonlight having the same color as sunlight.Space photography equipment choices
Debate over NASA's selection of the older Nikon D5 over newer models like the Z9, citing radiation resistance, proven reliability in space missions since 2017, and the addition of a Z9 for testing purposes on future Artemis missions.