Jul 26Monday, July 27, 2026 · all daysJul 28 · today »
1.Kimi-K3 on HuggingFace(huggingface.co)
1368 points by nateb2022 45 days ago | 538 comments | permalink
tl;dr: Moonshot AI released Kimi K3, an open-weight 2.8T-parameter MoE model (104B activated) with native multimodality and a 1M-token context window, built on new Kimi Delta Attention (KDA) and Attention Residuals architecture with MXFP4 quantization-aware training. Benchmarks show it competitive with Claude Opus 4.8, GPT-5.6, and Claude Fable 5 across reasoning, coding, agentic, and vision tasks, positioning it as the first open 3T-class frontier model. Weights are available on HuggingFace under the Kimi K3 License, with vLLM/SGLang deployment support and an OpenAI/Anthropic-compatible API.
HN Discussion:
  • Hosting costs and pricing dynamics for a 3T model will be revealing but expensive
  • Consumer hardware is poorly shaped for running large models locally
  • The real value is customization, fine-tuning, and IP sovereignty from open weights
  • License restrictions and revenue thresholds limit true openness of the release
  • Model has identity confusion claiming to be Claude, casting doubt on originality
2.PGSimCity - How PostgreSQL Works(nikolays.github.io)
915 points by jonbaer 45 days ago | 89 comments | permalink
tl;dr: PGSimCity is an educational, SimCity-styled interactive visualization that models PostgreSQL's internal engine to help users understand how it works. It's an early, unreviewed prototype that may contain inaccuracies, and contributions via issues or pull requests are welcomed. The project is independent and not affiliated with Electronic Arts.
HN Discussion:
  • ~Concept is appreciated but visualization is too busy and confusing to be effective
  • ~Would benefit from being interactive, e.g. entering a query and walking through the flow
  • Excellent educational approach that could be applied to other complex technical domains
  • Skeptical of accuracy since it was vibe-coded quickly with AI assistance
  • AI-assisted learning tools like this are a great new way to explore complex topics
3.Show HN: Physically accurate black hole you can put in your room(blackhole.plav.in)
472 points by aplavin 49 days ago | 183 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • Enthusiastic praise for the visualization's quality and smoothness
  • Critique that the 'physically accurate' claim is misleading science communication
  • ~Technical questions about specific accuracy details like accretion disk brightness asymmetry
  • Reports of technical issues with AR mode on certain devices
  • Tangential reflections on VR experiences and phobias triggered by large objects
4.French firefighters face 'pyrocumulonimbus' for first time(france24.com)
460 points by saaaaaam 45 days ago | 367 comments | permalink
tl;dr: French firefighters are battling a pyrocumulonimbus ("fire cloud") for the first time in France—a phenomenon previously confined to Australia and North America, where extreme ground heat drives a convective column that generates its own winds, lightning, and unpredictable multi-directional fire fronts. Officials say the blaze cannot be fought directly and have declared "operational impossibility," resorting to defensive tactics and hoping either for prolonged heavy rain or to steer the fire toward the sea.
HN Discussion:
  • Article's 'first time' claim is inaccurate per French sources citing prior events
  • Terminology quibble: should be pyrocumulus not pyrocumulonimbus since fire clouds don't rain
  • Background context on why the Landes forest is uniquely vulnerable to fire
  • Firsthand accounts and comparisons confirming the fire's severity and phenomenon
  • Broader climate crisis reflections, calling for geoengineering or noting widespread fires
5.Decker, a platform that builds on the legacy of Hypercard and classic macOS(beyondloom.com)
374 points by tosh 45 days ago | 99 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • Nostalgic appreciation for HyperCard's original extraordinary user experience
  • Questions whether self-contained app builders like this still have a place today
  • Project is nostalgic but impractical for real modern use
  • ~Excited about the tool's potential but wishes it were less quirky/more modernized
  • Relates Decker to similar tools like Lazarus, LiveCode, or tldraw
6.US citizen charged after GrapheneOS phone wipes during airport search(techspot.com)
1264 points by eecc 45 days ago | 1014 comments | permalink
tl;dr: Federal prosecutors in Atlanta are charging Sam Tunick with destroying property to prevent seizure after his GrapheneOS-equipped Pixel phone wiped itself during a border search at Hartsfield-Jackson airport. Experts say it may be the first time US law has been applied against use of a privacy-focused operating system, raising concerns that the case effectively criminalizes GrapheneOS by default. Tunick's defense argues agents denied him counsel and used a child sexual abuse material pretext to investigate his ties to the "Cop City" protest movement.
HN Discussion:
  • Border agents have excessive power; travelers need better threat modeling for device searches
  • Users who enter duress PINs must accept the legal consequences of intent-based laws
  • Article is sensationalist; defendant was charged for his actions, not for using GrapheneOS
  • The statute cited doesn't actually apply since the device was being searched, not seized
  • Better technical approaches (plausible deniability, remote access, blank phones) should be used instead
7.Htmx 4.0, the first JavaScript library to release exclusively on the Game Boy(swag.htmx.org)
500 points by rcy 46 days ago | 193 comments | permalink
tl;dr: Htmx 4.0 is being released as a Game Boy cartridge ($35.97 CAD), framed as a joke product where players collect pickles across four levels and must defeat "Warren" to unlock the htmx 4.0 source code. It's a novelty merch item rather than an actual software release.
HN Discussion:
  • HTMX's playful vibe and craft-focused culture make the Game Boy stunt a delightful reflection of the team
  • HTMX has genuinely transformed web development workflows for the better
  • ~HTMX is essentially a refined reinvention of older tech like .NET Web Forms update panels
  • HTMX has drawbacks like poor noscript fallbacks and encourages overuse on static sites
  • Curious whether the ROM will be available digitally beyond the physical cartridge
8.Design is compromise(stephango.com)
278 points by ankitg12 46 days ago | 89 comments | permalink
tl;dr: Compromise isn't a dirty word—it's just decision-making, and claims of "uncompromising" products are marketing nonsense since every design choice inherently trades one capability for another. The best products are opinionated, deliberately sacrificing broad appeal to excel at specific things. Good design means picking the right tradeoffs for your audience, not pretending tradeoffs don't exist.
HN Discussion:
  • Wholeheartedly agrees that compromise is a valuable, misunderstood skill in design and life
  • Compromise and tradeoff are not synonymous; the article conflates distinct concepts
  • ~Compromise should be a last resort after exhausting attempts to properly scope the problem
  • Good design transcends tradeoffs by layering or reframing rather than settling for compromise
  • ~Constraints aren't fixed; innovation can shift the tradeoff space itself
9.Show HN: Reverse Minesweeper(sunflowersgame.com)
252 points by pompomsheep 46 days ago | 87 comments | permalink
tl;dr: Reverse Minesweeper is a Minesweeper variant with a difficulty system driven by a logic engine that solves each generated grid using the simplest available deduction, classifying puzzles into five tiers (Easy, Medium, Hard, Extreme, Insane) based on the hardest technique required—from single-clue deductions up to Sudoku-style forcing chains and "either-way" coloring. Sliders let players tune minimum deduction counts and chain depths per tier, and a 0–100 score positions grids within non-overlapping difficulty bands, with estimated solve times computed from per-technique time weights.
HN Discussion:
  • Game is a nice, enjoyable take on Minesweeper worth returning to
  • This isn't novel — similar puzzle games (Mosaic, Fill-a-Pix, puzzlemadness) already exist
  • ~UX issues need fixing: single-click default action, dark theme, unclear naming as 'reverse'
  • The difficulty/logic engine falls short — some puzzles seem to require guessing or lack sufficient info
  • Curiosity about the author's other games and portfolio
10.Go Analysis Framework: modular static analysis by go team(pkg.go.dev)
218 points by AbuAssar 46 days ago | 89 comments | permalink
tl;dr: Go's `analysis` package defines a standard interface for modular static analyzers, where each `Analyzer` exposes a `Run` function operating on a `Pass` (syntax trees, type info, etc.) and reports diagnostics. Analyzers can depend on results from other analyzers and exchange serializable "Facts" across package boundaries—enabling separate-compilation-style analysis (e.g., printf-wrapper detection). The common interface lets tools like `vet`, IDEs, and build systems reuse checkers, with helper subpackages (`singlechecker`, `multichecker`, `analysistest`) for building and testing standalone commands.
HN Discussion:
  • General enthusiasm for Go's language design and its built-in tooling/linting philosophy
  • Practical endorsement: the framework enables teams to codify review knowledge into custom linters
  • Skepticism about the submission's novelty, noting the framework is already widely used
  • Curiosity about extending the framework to higher-level architectural linting
  • Appreciation for Go team's tooling focus as beneficial for both human and AI-assisted development
11.Kill The Cookie Banner(killthecookiebanner.eu)
1121 points by rapnie 46 days ago | 575 comments | permalink
tl;dr: The EU Commission has proposed letting users set privacy preferences once in their browser via automated signals, eliminating cookie banners—which noyb argues were designed to trick users into consenting (90% say yes, though only ~3% actually want tracking). Google and the tracking industry are lobbying EU Member States and Parliament to block the proposal. The provision is part of the broader "Digital Omnibus" reform, most of which noyb opposes as harmful to privacy rights.
HN Discussion:
  • Cookie banners cannot constitute informed consent and should be legally invalidated
  • Browser-based privacy preferences are an obvious solution that should extend to other regulations too
  • ~Browsers should isolate cookies by default rather than relying on consent mechanisms
  • ~The proposal doesn't go far enough — tracking itself should stop, or laws should mandate action like California did
  • Skepticism about the 3% tracking-consent statistic, suggesting survey misunderstanding
12.I learned PCB design, 3D printing and C just to listen to music(pentaton.app)
227 points by interfeco 49 days ago | 52 comments | permalink
tl;dr: Frustrated by the loss of album artwork in the streaming era, the author built "Pentaton LP," a 12"x12" AirPlay streamer featuring a 17" 1920x1920 IPS display designed to mimic a vinyl sleeve on display. The project required learning PCB design (a custom carrier board for a Radxa CM3), FreeCAD modeling, 3D printing, and low-level Linux work, plus a custom protocol extension to bypass AirPlay's low-resolution (500x500) artwork limitation. The device idles under 2W, wakes on streaming, and triggers the amplifier automatically; a Kickstarter is being considered.
HN Discussion:
  • Admiration for the project with curiosity about specific technical details or challenges
  • Enthusiasm for the concept and interest in buying or expanding features like lyrics
  • Questions about learning hardware skills and sourcing components as a hobbyist
  • Suggests simpler alternative solutions like a Mini PC with foobar2000 would suffice
  • Tangential reflection on PCB manufacturing accessibility and local fabrication shops
13.The New AI Superpowers: Focus and Followthrough(rickmanelius.com)
214 points by mooreds 46 days ago | 79 comments | permalink
tl;dr: AI productivity gains tempted the author to spin up 40+ side projects simultaneously, leading to burnout from self-imposed "make-work" rather than the promised freedom. The lesson: use AI to go deeper (vertical) on fewer things that matter, not wider (horizontal) across more projects—paying the costly "last 1%" that separates good-enough work from great work, à la a partial vs. total eclipse.
HN Discussion:
  • AI enables everyone to build redundant, incompatible versions of the same beginner software
  • Burnout stems from meaningfulness and confidence, not workload amount as article implies
  • AI coding agents actually reduce stress and burnout by flattening task difficulty
  • ~Managing AI like a team of contractors with proper process avoids burnout
  • The 100x productivity premise is a myth; AI output needs costly polishing to ship
14.What is happening to jobs? Separating AI hype from reality(siepr.stanford.edu)
299 points by pod_krad 46 days ago | 380 comments | permalink
tl;dr: Aggregate employment data shows little evidence of an AI-driven jobs apocalypse, though entry-level white-collar workers—especially in software and customer service—appear to be experiencing weaker hiring since 2022, with causes hard to disentangle from post-pandemic hiring corrections and interest rate hikes. Experimental studies show AI generally boosts worker productivity (particularly for less-skilled workers), but gains are uneven and haven't shown up in aggregate statistics. Firm adoption is accelerating but concentrated in tech and information-heavy sectors, with most companies still in pilot phases—suggesting AI's labor market impact, while real in pockets, remains early and uncertain.
HN Discussion:
  • Study data is outdated because coding/general agents only became effective in late 2024/2025
  • ~AI productivity gains are concentrated among top performers, amplifying existing inequality
  • AI helps less experienced workers but hinders experienced ones, matching article's claim
  • Organizational inertia masks real AI impacts that are unevenly distributed
  • ~Job market and AI hype from executives are disconnected from actual fundamentals
15.London Gatwick has launched a robotic airport parking service(aerospaceglobalnews.com)
288 points by agotterer 46 days ago | 259 comments | permalink
tl;dr: London Gatwick has partnered with Stanley Robotics to launch the UK's first robotic airport parking service, where drivers leave their car in an enclosed cabin and an autonomous robot lifts and transports it to storage—without requiring keys to be handed over. The system, bookable ahead of an August launch, increases parking density by packing cars closer together and retrieves vehicles based on customers' return flight details. It supports standard cars up to 2.6 tonnes and 2.3m tall.
HN Discussion:
  • Disappointed the service still requires a bus to the terminal rather than curbside drop-off
  • Excited to try the service and finds it genuinely cool despite being new
  • Pricing is reasonable for airport parking and makes driving to Gatwick viable
  • ~Questions practical issues like car alarms, key retention contradictions, and item retrieval
  • Skeptical the service will succeed, citing similar failed robotic parking ventures
16.GrapheneOS protections against data extraction from locked devices(discuss.grapheneos.org)
437 points by Cider9986 46 days ago | 251 comments | permalink
tl;dr: Summary not available.
HN Discussion:
  • GrapheneOS provides strong protection including BFU mode and auto-reboot features that help protect users
  • ~GrapheneOS needs additional features like backup/restore solutions for border crossings
  • ~Duress PIN should present a convincing fake OS rather than obvious wipe, or wipe specific apps invisibly
  • Pattern locks and short passwords provide insufficient entropy for real security
  • Simply powering devices off before border crossings is sufficient protection
17.The Strongest El Niño Ever(theclimatebrink.com)
282 points by ndsipa_pomu 45 days ago | 335 comments | permalink
tl;dr: Seasonal forecast models across 14 systems and 667 ensemble members are converging on a 2026-27 El Niño peak of ~3.6°C in the Niño 3.4 region — roughly 0.8°C above the 2015-16 record, with ~91% of members exceeding that prior high. Daily SSTs are already ~2°C above era-adjusted averages in mid-July, far outpacing 1997 and 2015 at this stage, though the author cautions that models have never been verified against an event this extreme. Most warming impact would hit 2027, likely making it the warmest year on record.
HN Discussion:
  • Highlights the alarming 2027 warming implications and urges personal preparation like AC/solar
  • Concerned that models underestimate ocean temps, signaling unpredictable extreme weather ahead
  • ~Frustrated that forecasts don't translate into actionable local predictions for individuals
  • Asks clarifying questions about what El Niño consequences actually mean locally (floods, hurricanes)
  • Fatalistic view that emissions won't be curbed, so only bracing or carbon capture remains
18.It's not empowering to hand off the details(davidnicholaswilliams.com)
222 points by davnicwil 45 days ago | 142 comments | permalink
tl;dr: AI enthusiasm is often driven by a fantasy of achieving results without engaging with the messy details, but expertise and quality work inherently require deep, meticulous attention to those details. You can only successfully delegate parts of a task if you're already skilled enough to know which parts—meaning AI can't shortcut the process of becoming good at something. Handing off the details entirely isn't empowerment; it's the opposite, since it means you've contributed nothing.
HN Discussion:
  • ~Judgment about which details to delegate is a learned skill that AI use still requires
  • Abstraction and delegation are normal in every field; panic about AI is overblown
  • AI lets me offload boring boilerplate and focus on details I actually care about
  • Verification doesn't require full understanding, so delegation isn't a loss of agency
  • Skipping the hard parts with AI stunts growth, just like shortcuts in the gym
19.Android may soon restrict on-device ADB(kitsumed.github.io)
998 points by shscs911 47 days ago | 499 comments | permalink
tl;dr: Google is considering restricting Android's ADB daemon to only bind to the wlan0 interface in response to CVE-2026-0073, which would break on-device ADB (loopback connections used by tools like Shizuku, App Manager, and Canta) along with ADB over VPN or Ethernet. The author argues this framing treats legitimate on-device ADB use as an exploit vector, when in practice attackers can't start ADBD themselves, and proposes a persistent user-toggleable setting instead of a hard block.
HN Discussion:
  • The attack vector is unrealistic and doesn't justify breaking legitimate ADB use cases
  • This is part of Google's broader trend of locking down Android and eroding openness
  • Android losing openness eliminates the main reason to choose it over iOS
  • Restricting ADB to specific interfaces like VPN would actually be a security improvement
  • Users need alternatives like Linux phones since Android is becoming untrustworthy
20.Show HN: I mapped every US golf course(golfcoursebrowser.com)
216 points by rickmf 46 days ago | 166 comments | permalink
tl;dr: Summary not available.
HN Discussion:
  • ~Data should be contributed back to OpenStreetMap for community benefit and sustainability
  • Project risks legal liability by scraping others' data without originality
  • Suggestions to enhance the map with polygons, details, and mini golf courses
  • Golf courses are an exclusionary misuse of public/urban land resources
  • The map is fascinating for exploring geographic and cultural patterns of golf