HackerNews Digest

April 03, 2026

Google releases Gemma 4 open models

Gemma 4 is a Google DeepMind platform for creating autonomous agents capable of planning, navigating applications, and executing tasks on users’ behalf. It includes built‑in support for function calling, enabling agents to invoke external APIs or services directly. The page also contains two footer images, labeled “footer_gemma__light” and “footer_gemma__dark.”
Read full article →
Comments show a mixed but overall favorable view of Gemma‑4. Reviewers highlight strong performance on desktop GPUs, fast token generation, useful open‑source licensing, and promising multimodal and audio features, while noting that the 31B dense model still lags behind Qwen 3.5 on many benchmarks. Recurrent concerns include tool‑calling inconsistencies, broken or slow behavior on low‑end hardware, confusing “effective” parameter naming, and the lack of documented features such as TurboQuant. Many users appreciate community tooling and quantization efforts, yet several call for clearer benchmarks, better documentation, and more reliable deployment pipelines.
Read all comments →

Decisions that eroded trust in Azure – by a former Azure Core engineer

None
Read full article →
The comments converge on a broadly critical view of Azure, describing the UI as disjointed, documentation frequently outdated, and the service portfolio overwhelming without clear guidance. Multiple contributors cite reliability problems, difficult provisioning, and inadequate support, often comparing it unfavorably to competing clouds. Several remarks highlight internal organizational challenges, chronic understaffing, high turnover, and burnout among engineers. A minority acknowledge that large‑scale platforms inevitably have rough edges and that Azure continues to operate, but the prevailing sentiment emphasizes systemic flaws and user frustration.
Read all comments →

The open web isn't dying. We're killing it

The essay argues that the open web’s decline is not a new phenomenon caused solely by AI, but the result of long‑standing market dynamics and collective user choices. Large platforms gained power because users repeatedly migrated to them for convenience—easier identity, payments, discovery, hosting, moderation, and analytics—despite knowing that “free” services rely on advertising‑driven surveillance and centralization. This convenience compounded, making openness appear costly and niche. The author stresses that neglect, not innocence, contributed to the shift: users, organizations, and policymakers favored short‑term consumer surplus over sustained economic support for publishing tools, independent hosting, RSS, and moderation. To revive an open web, the piece calls for cultural change—paying for services, using exportable tools, supporting independent software, and rebuilding portable social and economic primitives—so participants act as maintainers and contributors rather than passive consumers. The goal is to create resilient, portable systems that preserve user agency and freedom of movement.
Read full article →
The comment expresses a bleak view of the open web, arguing that most users are indifferent, prioritizing mobile shopping and social media over supporting small sites. It criticizes the tech community for focusing on AI press releases and neglecting personal encouragement for creators, suggesting systemic indifference outweighs individual effort. The author doubts that personal sacrifice or better usability alone can revive the web, viewing current trends and declining software quality as signs that the open web is effectively fading.
Read all comments →

Tailscale's new macOS home

Tailscale’s original macOS client combined a command‑line tool with a menu‑bar status item. On MacBook models with a display notch (starting 2021), the status‑item icon can be rendered within the notch’s invisible zone when other menu‑bar icons fill the available space, making the icon inaccessible. macOS provides no native overflow or repositioning for third‑party items; Apple’s indirect work‑arounds (e.g., moving system icons to Control Center) and third‑party managers such as Bartender are the only options. Tailscale addressed the problem by observing the window’s occlusionState via NSWindow.didChangeOcclusionStateNotification; when the icon is not visible, the app displays a pop‑up warning. This solution is imperfect and may trigger on lid or monitor changes. A more robust fix arrived with the windowed macOS client (enabled by default in version 1.96.2), which runs alongside the menu‑bar app and offers: - searchable device list with status, - ping, IP copy, and Taildrop file transfer, - exit‑node selection with latency‑based recommendation, - Dock‑icon error indicator, - optional “mini player” mode and onboarding tour. The windowed UI reduces reliance on the notch‑prone menu‑bar icon.
Read full article →
Users express strong frustration with macOS’s notch obscuring menu‑bar icons, citing lost sales, support headaches, and the lack of a native overflow solution, while adopting manual defaults tweaks or third‑party managers despite recent instability. Opinions about Apple’s design choices are uniformly critical, urging a redesign or overflow dropdown. Tailscale receives mixed feedback: many praise its ease of setup and reliability, yet complain about UI changes, SSO complications, occasional packet loss, and the need for CLI alternatives. Overall, there is a clear demand for better system UI handling and more stable, user‑friendly networking tools.
Read all comments →

C89cc.sh – standalone C89/ELF64 compiler in pure portable shell

The repository “c89cc.sh” provides a self‑contained C89 compiler that generates ELF64 binaries using only portable shell scripts, without external toolchains. It implements the C89 language front‑end, parsing, code generation, and ELF64 output entirely within POSIX‑compatible shell code, enabling compilation on systems lacking a native compiler. The project includes example usage and documentation, and the GitHub page features contributor avatars labeled “@alganet” and “@lucaraymaekers”. The tool is positioned as a minimalistic, pure‑shell solution for building C programs in constrained environments.
Read full article →
The remarks show cautious interest in the script, with a desire for safer execution through added tests, documentation, and a maintained repository rather than a simple gist. Users note its single‑file, tool‑free design, empty PATH, and claimed portability across several shells while producing x86 ELF binaries, yet there is confusion about whether it targets x86‑64 ELF or a generic shell. Despite uncertainty, the code’s apparent auto‑generation is widely regarded as impressively sophisticated. Overall, the feedback balances admiration with requests for better packaging and verification.
Read all comments →

Artemis II's toilet is a moon mission milestone

NASA’s Artemis II will be the first crewed lunar mission to fly a functional toilet, the Universal Waste Management System (UWMS). Developed by Collins Aerospace under a NASA contract since 2015, the UWMS replaces Apollo‑era bag‑and‑funnel kits with a 3‑D‑printed titanium unit that handles urine and feces simultaneously, provides a solid door for privacy, and includes handholds for microgravity stability. The system is unisex, uses vacuum‑assisted collection, and can be adapted for future lunar and Mars habitats. A prototype was tested on the International Space Station in 2020, with final ISS installation in 2021; a lunar‑specific version has been installed in Orion for Artemis II. NASA officials cite the toilet as “mission‑critical,” noting that earlier Apollo waste systems were unreliable, prone to leaks, and caused crew dissatisfaction. Successful operation on Artemis II is intended to inform waste‑management designs for subsequent Artemis flights and long‑duration deep‑space missions.
Read full article →
Comments acknowledge that early space‑flight waste systems were technically functional but uncomfortable, time‑consuming, and prone to spills, leading to crew dissatisfaction and extensive troubleshooting. The discussion notes gender‑related usability issues, the complexity of handling solid and liquid waste, and the humor surrounding the language used by astronauts and engineers. Recent Artemis II problems are cited as evidence that plumbing challenges persist, while improvements such as combined solid‑liquid handling and design tweaks receive praise. Overall, the tone blends amusement, recognition of engineering difficulty, and a desire for clearer technical explanations.
Read all comments →

Cursor 3

Cursor 3 is a redesigned, agent‑centric IDE that unifies local and cloud development workflows. The interface supports multi‑workspace projects, letting humans and AI agents operate across multiple repositories simultaneously. All agents—local, cloud, and those triggered from mobile, web, Slack, GitHub, or Linear—appear in a shared sidebar; cloud agents generate visual demos for verification. Users can transfer an agent session between cloud and desktop instantly, enabling on‑device edits with Composer 2 (a high‑limit coding model) or background execution when offline. A new diffs view streamlines staging, committing, and PR management. Additional features include full LSP navigation, an integrated browser for local site interaction, and a marketplace offering hundreds of plugins (MCPs, skills, subagents) with private team options. Cursor 3 aims to provide the core model, product, and runtime needed for more autonomous agents and collaborative AI‑enhanced coding, while retaining a conventional IDE fallback.
Read full article →
Comments show mixed reactions to Cursor’s recent direction. Many users criticize the shift toward a chat‑first, agent‑centric UI, feeling it diminishes code‑centric workflows, adds unnecessary complexity, and replicates designs of competing tools, while also raising cost and performance concerns. Some appreciate the ability to switch models, integrated cloud agents, and occasional productivity gains, especially for large codebases, but a sizable portion plans to revert to traditional IDEs or alternative platforms, citing better value, familiarity, and more effective developer control. Overall sentiment leans toward disappointment with the new approach.
Read all comments →

Qwen3.6-Plus: Towards real world agents

None
Read full article →
Comments show a mixed view of Qwen’s latest closed‑weight models. Users acknowledge solid benchmark scores and see value in a cheaper, non‑SOTA option, especially when free token limits are available. Criticism centers on the practice of comparing against older competitor versions, which many deem misleading, and on occasional hallucinations and tool‑call loops. Some express interest in API control and long‑horizon agent tasks, while others remain wary of privacy and prefer open‑weight or alternative providers. Overall sentiment balances appreciation for progress with skepticism about marketing tactics and closed‑source constraints.
Read all comments →

Good ideas do not need lots of lies in order to gain public acceptance (2008)

The May 27 2004 “D‑squared Digest” entry links business‑school concepts to political judgment, especially regarding the Iraq war. The author recounts an accounting class debate on expensing stock options, concluding that genuinely beneficial ideas would be fully disclosed, while attempts to hide costs indicate falsehood. Applying this to Iraq, the post argues that officials’ claims about weapons of mass destruction were dishonest, so their forecasts should be dismissed entirely. It stresses that projects driven by untrustworthy forecasts—without post‑project audits—inevitably fail, and that audit culture is essential to prevent repeated reliance on liars. The piece references Paul Krugman, the Paulson bailout plan, the book *Development, Geography and Economic Theory*, and suggests using Benford’s Law to test war‑justification data. Overall, it presents a concise argument: good ideas require transparency, and granting credibility to known liars undermines sound decision‑making.
Read full article →
The comments converge on a skeptical view of deception in promoting ideas, noting that while falsehoods can temporarily mobilize support for otherwise beneficial changes, persistent lying signals a flawed concept. Examples range from tech stock‑option adoption and AI hype to climate action and the Iraq war, illustrating how persuasive marketing often outweighs factual merit. Contributors stress that public acceptance is an unreliable gauge of truth, advocate for honest framing and critical analysis, and warn that unchecked hype and herd behavior can undermine sound decision‑making.
Read all comments →

Post Mortem: axios NPM supply chain compromise

In March 2026 two malicious Axios releases (1.14.1 and 0.30.4) were published to npm from a compromised maintainer account. Both versions introduced the dependency **plain‑crypto‑[email protected]**, which installed a remote‑access trojan on macOS, Windows and Linux. The packages were live for roughly three hours before removal. **Impact & remediation** - Search lockfiles for `[email protected]|0.30.4` or `plain-crypto-js`; if found, treat the host as compromised. - Downgrade to `[email protected]` (or `0.30.3` for 0.x), delete `node_modules/plain-crypto-js`, rotate all secrets, and audit network traffic for connections to `sfrclak.com` or `142.11.206.73:8000`. - Apply the same steps on CI runners and rotate any injected credentials. **Root cause** - Targeted social‑engineering and RAT malware gave the attacker access to the lead maintainer’s PC and npm credentials, allowing unauthorized publishes. **Response** - Full wipe of the maintainer’s devices and reset of all related accounts. - Deprecation of the compromised versions, removal of `plain-crypto-js` from npm, and coordination with npm security. - Planned security upgrades: OIDC publishing flow, immutable releases, automated publish monitoring, and tighter maintainer access controls.
Read full article →
The comments convey concern about a recent supply‑chain breach, noting that the malicious release lacked OIDC provenance attestations while legitimate versions included them, and highlighting that such discrepancies often go unchecked. Observers point to a broader rise in similar attacks and criticize npm’s current security measures for insufficient detection of malicious code in public packages. There is interest in detailed payload analysis to assess the breach’s scope, and questions about whether OIDC flows could mitigate this type of intrusion.
Read all comments →