Top 7 Hacker News posts, summarized
HN discussion
(1402 points, 2420 comments)
Apple has announced the iPhone Duo, its first foldable iPhone featuring a 7.6-inch inner Super Retina XDR display (50% larger than the iPhone 18 Pro Max) and a 5.4-inch outer display with consistent aspect ratios. The device uses a Grade 5 titanium frame and hinge cover, Ceramic Shield front and back, a scratch-resistant coating on the inner display, and IP68 water/dust resistance. It runs on the A20 Pro chip with a Dual 16-core Neural Engine, an iPhone-first dual-battery system, eSIM-only design, and the C2 modem. iOS 27 introduces reimagined experiences for multiple poses — landscape, portrait, closed, seated, and standing — enabling Split View multitasking, hands-free FaceTime, and flexible viewing angles. The camera system includes a 48MP Dual Fusion (Main + Ultra Wide), 12MP Center Stage front camera, and an under-display FaceTime camera, with unique features: Smart Take, Duo Preview, Kid Cue, and Duo FaceTime. Authentication uses Touch ID in the side button (no Face ID). Apple Pencil (USB-C) support is coming later. The iPhone Duo starts at $1,999 with 256GB, 512GB, 1TB, and 2TB storage options in Night Sky and Star White, available October 23, 2026. Environmental claims include 35% recycled material by weight, 85% recycled titanium, 60% renewable electricity manufacturing, and 100% fiber-based packaging.
HN commenters expressed skepticism about the foldable form factor's practicality and durability, with several questioning real-world demand and noting the $1,999 price matches Samsung's foldables. The eSIM-only design drew criticism for limited global compatibility. The absence of Face ID (replaced by side-button Touch ID) was a dealbreaker for some users. Positive notes included Apple Pencil support — seen as a potential threat to devices like reMarkable — and optimism that Apple may have solved hinge and crease issues. The outer display's 5.4-inch size was compared to the iPhone 13 mini. Many commenters adopted a wait-and-see approach, citing first-generation risks (referencing Vision Pro) and wanting confirmation of long-term hardware reliability and software commitment before purchasing. A few noted the design resembles Microsoft's Surface Duo.
HN discussion
(640 points, 433 comments)
Shopify announced it is migrating its mobile apps (Shopify, Shop, Point of Sale, Inbox) from React Native back to native Swift and Kotlin, reversing a successful 2020 decision. The original move to React Native aimed to avoid building features twice, enable cross-stack contributions, and reduce feature-parity overhead. While React Native delivered these benefits, Shopify states that the rise of advanced coding agents (LLMs) has fundamentally altered the cost equation: agents can now implement, translate, test, and review features across iOS and Android autonomously, dramatically reducing the marginal cost of maintaining two native codebases. Shopify prototyped rebuilding core app sections with agents and found the process surprisingly effective. They are pursuing a greenfield rebuild strategy—starting with the Shop app, which shipped in 12 weeks—assisted by an internal system called Helix that enforces quality through incremental checkpoints, automated tests, visual reviews, adversarial AI reviewers, and human approval. They also architected a headless CLI layer to give agents fast feedback loops without slow simulator interaction. Shopify will wind down its React Native open-source libraries (Skia sponsored through 2026, FlashList seeking new stewards, Restyle archived) but emphasized this is a context-specific decision driven by their scale and agent infrastructure, not a blanket indictment of React Native.
The HN discussion revealed deep skepticism about whether AI truly changes the long-term economics of cross-platform development. The most upvoted critique argued that Shopify's post demonstrates only that rewrites are cheaper with AI, not that ongoing parity maintenance is sustainable, noting that feature divergence accumulates over years through experiments, edge cases, and platform changes—areas where agents currently struggle. Several commenters questioned the lack of concrete cost data (engineer hours, token spend, defect rates) for a decision framed as a first-principles reevaluation. Others pointed out that the agent-friendly architecture (headless CLI, decoupled business logic) could benefit React Native equally, making the native rewrite seem like a solution to a tooling problem rather than a framework limitation. Some developers shared personal success stories using agents to rewrite smaller apps to native overnight, while others countered that React Native's hot reload and Expo's agentic tooling remain superior for iteration speed. A recurring theme was concern for the ecosystem impact of Shopify abandoning stewardship of critical libraries like FlashList and Skia. A minority viewed this as validation of a broader industry trend toward native, while others dismissed it as a recurring fashion cycle or a decision influenced by leadership preferences rather than technical necessity.
HN discussion
(333 points, 162 comments)
Windows XP selected initial user account pictures using a one-pass reservoir sampling algorithm (with k=1) implemented via `RtlRandomEx` seeded by `GetTickCount()`. The algorithm iterates through pictures in the Default Pictures directory, maintaining a current "winner" that gets replaced with probability 1/n at the nth item. This approach avoids two filesystem passes (counting then selecting) and handles directory changes during execution. As a safeguard against pathological cases, sampling stops after 100 pictures. The implementation is a special case of reservoir sampling that permits a simpler tailored algorithm.
Commenters expressed nostalgia for XP's default pictures and debated the 100-file limit (some calling it unnecessarily low). Several noted the chessboard image often associated with the Administrator account. The discussion highlighted appreciation for Raymond Chen's long-running Windows internals blog, with a link shared to the actual source code from leaked Windows XP sources. Technical discussion compared reservoir sampling to naive approaches (array storage with random index) and alternative algorithms (random key assignment), while noting filesystem limitations like lack of a constant-time file count operation. Some questioned why username wasn't used as a seed for deterministic per-user assignment.
HN discussion
(332 points, 136 comments)
Cognition has launched SWE-2, a new coding model post-trained from Moonshot AI's Kimi K3 (2.8T parameters) that achieves 50.0% on FrontierCode 1.1 Main—within one point of Fable 5.1 while being 64% cheaper. The model beats SWE-1.7 and Grok 4.6 on both score and cost, matches GPT-5.6 Sol and Fable 5/5.1 at a fraction of their price, and comes within a few points of GPT-6 Astra at roughly a quarter of the cost. Key technical advances include a Pareto-informed cost-penalty RL algorithm that trains all reasoning-effort levels in a single run, a length-weighted reward baseline that stabilizes training, improved RL rollout serving with NVFP4/FP8 kernels and online draft-model training, and a data flywheel that tripled RL environments and hardened verifiers against reward hacking. Behaviorally, SWE-2 shows 58% fewer turns and 81% lower cost on FrontierCode, with focused exploration (first edit at median 18 steps vs. 48 for SWE-1.7), improved test coverage, resourcefulness, and verification discipline. The model is available now in Devin Desktop, CLI, Web, and Fusion. Trustworthiness evaluations show a 98% pass rate on propaganda/censorship tests and no statistically significant framing effects on context-dependent vulnerability.
HN commenters express significant skepticism about SWE-2's benchmark claims, noting a massive delta between Terminal-Bench 2.1 (92.8%) and the newer Terminal-Bench 4 (27.3%), suggesting potential benchmaxing on older evaluations. The fact that SWE-2 is a post-train of Kimi K3 rather than a new base model draws both praise for efficiency and criticism for lack of novelty. A major friction point is platform lock-in: SWE-2 requires Devin's proprietary ecosystem, which many users reject in favor of open-weight alternatives like DeepSeek Flash 4.1 or flexible tools like Codex and Gemini that integrate into existing workflows. Cognition's past credibility issues—specifically the controversial 2024 Devin demo—are cited as reasons to treat performance claims cautiously. Some commenters appreciate the technical transparency around Pareto-frontier optimization and RL infrastructure, while others question the relevance of propaganda/censorship evaluations for a coding model. A few users report strong real-world productivity gains with competing models (particularly Fable 5.1) for complex distributed systems work, contrasting with those who found SWE-1.7 unusably slow and prone to overthinking.
HN discussion
(340 points, 112 comments)
Four California plaintiffs filed a class-action lawsuit against Sony Interactive Entertainment in June 2026, alleging the PlayStation Store misleads consumers by using ownership language such as "Buy Now" and "Confirm Purchase" while granting only a revocable license. The complaint cites California's AB 2426 (effective January 2025), which prohibits advertising digital goods with terms implying unrestricted ownership unless the seller provides a clear, conspicuous disclosure that the transaction is a license, including all restrictions and revocation terms. Plaintiffs claim the disclosure appears in small, unhighlighted text above the purchase button, falling short of the statutory standard. Sony moved to compel individual arbitration under its Terms of Service, which include a class-action waiver and a 30-day opt-out provision that no plaintiff exercised. In its motion to dismiss, Sony argued that "reasonable consumers would not be misled" into believing they obtain ownership of a digital game, contending that because multiple users can purchase the same title, exclusive ownership is implausible. The article documents numerous instances where Sony's own support pages, marketing emails, and store listings refer to players "owning" digital games, purchasing add-ons for games they "already own," and verifying "license ownership" — language plaintiffs argue contradicts Sony's legal position. A hearing is scheduled for October 1, 2026.
Commenters widely criticized Sony's argument that digital ownership is impossible because multiple people can buy the same game, comparing it to physical media where each buyer owns a distinct copy. Several highlighted the arbitration clause and opt-out mechanism as effectively unenforceable for most users, calling binding arbitration a tool to strip consumer rights. Others emphasized the need for legal reform establishing true digital ownership, including escrow arrangements for continued access after store shutdowns, and advocated for DRM-free distribution. Some noted the hypocrisy of Sony using ownership terminology in its own communications while denying it in court. A few commenters questioned whether the focus on Sony reflects astroturfing, given that similar licensing practices are industry-wide, though most viewed the case as a pivotal test for digital consumer rights.
HN discussion
(245 points, 198 comments)
The author recounts discovering a social media post by Tyler Gaw advocating for keeping a "Big Box of Cables" after finding needed cables that had been stored for over a decade. Inspired, the author printed the post, taped it to their own cable box (labeled "FAMILY TECHNO BOX" by their spouse), and now uses it as a reminder of the box's value and a deterrent against family members discarding it. The author hopes the labeled box will someday pass on this philosophy to their children.
Commenters share varied approaches to cable management, ranging from single-bin minimalism to elaborate multi-box systems sorted by cable type (USB, audio, power, legacy standards like SCSI and FireWire). Many recommend clear zippered bags or Hefty slider bags for visibility and tangle prevention, while others describe purging obsolete proprietary cables and consolidating via USB-C adapters. Several note the scrap value of copper cables, and a few highlight professional needs requiring extensive inventories. The tension between "just in case" hoarding and deliberate curation recurs, with some advocating strict limits (e.g., 3-4 per cable type) and others celebrating the serendipitous utility of long-kept items.
HN discussion
(136 points, 258 comments)
The article examines anthropologist Roberto González's report on the growing integration of Silicon Valley and Big Tech into the U.S. military-industrial complex. The Pentagon is increasingly adopting AI-enabled systems, directing substantial funding to major tech corporations—Microsoft, Amazon, and Alphabet received approximately $28 billion combined from 2018 to 2022, with top five contract ceilings totaling at least $53 billion from 2019 to 2022. Venture capital investment in defense tech startups surged to nearly $100 billion from 2021 through 2023, a 40% increase over the prior seven years. González warns that classified contracts obscure true spending levels and that the resulting high-tech defense products may be ineffective, unpredictable, and unsafe, all funded by taxpayers.
Commenters emphasize that Silicon Valley's ties to the military are not new—historical examples include Fairchild Semiconductor's missile contracts, ARPANET, and CIA-backed In-Q-Tel funding Keyhole (later Google Earth). Many argue traditional prime contractors failed to deliver cost-effective solutions, creating an opening for VC-backed startups to disrupt the market. Ethical concerns arise: some tech workers refuse to build military systems, while others warn Silicon Valley's "move fast" culture neglects security, risking rapid reverse-engineering by adversaries. Political connections draw scrutiny, including Trump-family-owned Powerus and Kushner/Thiel/Andreessen-backed ventures like Anduril. Opinions diverge on whether this shift brings SpaceX-like efficiency or merely extracts tax dollars for surveillance and profiteering, with broader debate on whether companies should engage in defense work at all.
Generated with hn-summaries