| 1. | Einstein's relativity rules chemical bonds in heavy elements, new research shows(brown.edu) |
| 268 points by hhs 13 hours ago | 100 comments | permalink | |
tl;dr: Brown University chemists used photoelectron spectroscopy on carbon-bismuth molecules to show that triple bonds in heavy elements don't follow the textbook one-sigma-plus-two-pi structure. Due to relativistic effects (spin-orbit coupling) in heavy nuclei, the bonds instead appear as one pi bond and two hybrid sigma-pi bonds. The finding provides direct experimental evidence for a long-theorized relativistic bonding regime and could prompt textbook revisions as heavy elements like bismuth gain importance in solar cells and quantum computing research. | |
HN Discussion:
| |
| 2. | QuadRF can spot drones and see WiFi through my wall(jeffgeerling.com) |
| 606 points by speckx 19 hours ago | 201 comments | permalink | |
tl;dr: QuadRF is an open-source handheld phased-array SDR (4.9-6 GHz) built on a Raspberry Pi 5 and FPGA, streaming I/Q data over the Pi's MIPI lanes at 5+ Gbps. In hands-on testing it successfully visualized WiFi networks through walls and tracked a DJI drone in flight via an AR overlay, though the UI is still rough. It's a scaled-down piece of a larger project aiming to chain modules for Earth-Moon-Earth radio experiments, currently crowdfunding on Crowd Supply starting at $499. | |
HN Discussion:
| |
| 3. | Apple sues OpenAI, accuses ex-employees of stealing trade secrets(9to5mac.com) |
| 1135 points by stock_toaster 15 hours ago | 597 comments | permalink | |
tl;dr: Apple has sued OpenAI, io Products, and two former Apple employees—ex-VP Tang Tan and engineer Chang Liu—alleging they stole trade secrets to benefit OpenAI's hardware efforts led by Jony Ive. The complaint claims Tan solicited confidential info and Apple hardware samples from job candidates, Liu exploited a security bug to download thousands of pages of engineering files, and OpenAI tapped Apple suppliers using insider knowledge. Apple, which says over 400 ex-employees now work at OpenAI, is seeking damages and injunctive relief as OpenAI prepares its first consumer hardware device. | |
HN Discussion:
| |
| 4. | An update on residential proxies and the scraper situation(lwn.net) |
| 211 points by chmaynard 16 hours ago | 217 comments | permalink | |
tl;dr: AI scraper traffic increasingly comes through "residential proxy" networks—millions of compromised or opt-in devices (often via shady free VPNs, SDKs, or malware-infected streaming boxes) that make each request look like a unique human visitor, defeating traditional IP-based blocking. LWN has resisted deploying tools like Anubis to avoid burdening real users, instead using undisclosed optimizations and defenses, while noting that proof-of-work is a weak deterrent when attackers command millions of hijacked machines. Recent law-enforcement takedowns of networks like IPIDEA and NetNut have offered only temporary relief in an ongoing arms race threatening the open web. | |
HN Discussion:
| |
| 5. | Good Tools Are Invisible(gingerbill.org) |
| 452 points by theanonymousone 1 day ago | 209 comments | permalink | |
tl;dr: Good tools should disappear into the background, but users often mistake the friction of working around a tool's shortcomings (like crafting vim macros or endlessly configuring Linux) for genuine productivity, conflating cleverness with output. This tribal attachment turns tool choice into identity, causing people to defend and even celebrate flaws rather than acknowledge them. The responsibility lies with toolmakers to ship sensible defaults rather than offloading decisions onto users under the guise of "configurability." | |
HN Discussion:
| |
| 6. | AI 2040: Plan A(ai-2040.com) |
| 290 points by kschaul 1 day ago | 290 comments | permalink | |
tl;dr: The AI 2027 authors present "Plan A," a proposed positive alternative to their previous predictions of extinction or power concentration from superintelligence. The plan calls for delaying superintelligence development until 2040, making all AI research public, letting dozens of global companies reach the frontier, and establishing "mutually assured compute destruction." The scenario depicts 2027-2029 as a critical window where mass white-collar job disruption, looming recursive self-improvement, and concentration of power in a few US/Chinese CEOs force Congress and voters to confront who ultimately controls AI—culminating in a pivotal 2028 election. | |
HN Discussion:
| |
| 7. | Late Bronze Age Collapse(acoup.blog) |
| 377 points by dmonay 23 hours ago | 261 comments | permalink | |
tl;dr: The Late Bronze Age Collapse (c. 1220–1170 BC) was a wave of site destructions and state failures across the Eastern Mediterranean and Near East, hitting Mycenaean Greece and the Hittite Empire hardest while Egypt, Assyria, and Babylon declined but survived. The likely cause was a combination of factors—drought-driven harvest failures, intensifying warfare straining centralized palace economies, and cascading disruptions from refugees/raiders (the "Sea Peoples")—rather than any single cataclysm or migration like the debunked "Dorian Invasion." The aftermath saw Greece deurbanize and lose writing entirely, but the resulting fragmentation enabled the rise of the Phoenicians, the Greek polis, and the kingdoms of Israel and Judah. | |
HN Discussion:
| |
| 8. | The tech of 'Terminator 2' – an oral history (2017)(vfxblog.com) |
| 219 points by markus_zhang 19 hours ago | 78 comments | permalink | |
tl;dr: An oral history with ILM veterans on the pioneering CG tools built for Terminator 2's T-1000, including "Body Sock" (for stitching b-spline patches across animated joints), "Make Sticky" (an early UV-mapping technique for texture adherence), and a custom "poly alloy" RenderMan shader that faked ray-traced reflections using placed reflection planes. The team of ~12-15 people used Alias and SGI hardware, rotoscoped Robert Patrick from dual VistaVision cameras to build the digital character, and largely invented modern VFX workflows on the fly—including overnight rendering, match-move edge detection, and procedural displacement for shots like the head-through-bars. | |
HN Discussion:
| |
| 9. | GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture [pdf](cdn.openai.com) |
| 466 points by scrlk 17 hours ago | 376 comments | permalink | |
tl;dr: Summary not available | |
HN Discussion:
| |
| 10. | New York City to ban deceptive subscription practices(theguardian.com) |
| 537 points by randycupertino 17 hours ago | 255 comments | permalink | |
tl;dr: Starting October 1, New York City will fine companies $525 per subscription (plus back fees) if they don't offer a simple cancellation method, targeting gyms, streaming services, and other recurring charges. A separate proposed rule would require sellers—including landlords, hotels, and rental car agencies—to advertise total upfront prices inclusive of mandatory "junk fees," which would notably affect NYC's rental market where hidden fees inflate advertised rents. The moves follow the failure of similar Biden-era federal rules struck down in court or watered down by industry lobbying. | |
HN Discussion:
| |
| 11. | Show HN: Getting GLM 5.2 running on my slow computer(github.com) |
| 860 points by vforno 2 days ago | 214 comments | permalink | |
tl;dr: Colibrì is a dependency-free ~2,400-line C engine that runs the 744B-parameter GLM-5.2 MoE model on consumer hardware (~25 GB RAM) by keeping dense weights (~9.9 GB int4) resident while streaming the 21,504 routed experts from disk (~370 GB) via an LRU cache. It implements MLA attention with compressed KV-cache, DSA sparse attention, native MTP speculative decoding, KV persistence across sessions, and an optional CUDA tier for hot experts. Measured throughput ranges from 0.05 tok/s on the dev box's WSL2 VHDX to ~1 tok/s on an M5 Max, with faster NVMe/RAM shifting the bottleneck from disk to matmul. | |
HN Discussion:
| |
| 12. | Successful companies go blind(ianreppel.org) |
| 218 points by speckx 22 hours ago | 78 comments | permalink | |
tl;dr: Successful companies develop "competence blindness" like Mexican cavefish that lose their sight in caves: when market barriers protect incumbents, careful engineering becomes a vestigial trait, and employees who only know the internal culture perpetuate it through hiring. Attempts to fix this via "centres of excellence" actually suppress distributed excellence by centralizing it into bureaucratic process shops. Sighted engineers who join either leave quickly or gradually adapt to the dysfunction—making staying a form of apoptosis rather than loyalty. | |
HN Discussion:
| |
| 13. | Show HN: 18 Words(18words.com) |
| 1104 points by pompomsheep 1 day ago | 351 comments | permalink | |
tl;dr: Summary not available | |
HN Discussion:
| |
| 14. | GPT-5.6(openai.com) |
| 1524 points by logickkk1 1 day ago | 1086 comments | permalink | |
tl;dr: Summary not available | |
HN Discussion:
| |
| 15. | How the terrorist group Boko Haram uses frontier AI(casp.ac) |
| 207 points by imustachyou 17 hours ago | 173 comments | permalink | |
tl;dr: Summary not available | |
HN Discussion:
| |
| 16. | EU Parliament greenlights Chat Control 1.0(patrick-breyer.de) |
| 1596 points by rapnie 2 days ago | 826 comments | permalink | |
tl;dr: The EU Parliament failed to block "Chat Control 1.0," allowing suspicionless scanning of private messages on US platforms like Gmail, Instagram, and Discord to continue until 2028, despite a majority of voting MEPs (314 vs 276) opposing it—the rejection motion fell short of the required absolute majority of 361. End-to-end encrypted services like WhatsApp remain exempt. Negotiations over a permanent "Chat Control 2.0" resume in September, with Parliament pushing for targeted, court-ordered detection instead of blanket scanning, while member states favor maintaining the status quo. | |
HN Discussion:
| |
| 17. | Apple Silicon Exec Explains Mac Mini AI Demand and On-Device Future(macrumors.com) |
| 203 points by tosh 4 days ago | 293 comments | permalink | |
tl;dr: Apple's Doug Brooks says the Mac mini and Mac Studio are seeing strong demand for running AI agents, as developers want isolated, always-on systems separate from their primary machines. He frames agentic AI as a whole-chip workload—leveraging the Neural Engine, CPU neural accelerators, and newer GPU accelerators—rather than a pure GPU task, and predicts a hybrid future where agents dynamically split work between on-device and cloud inference driven by privacy and rising token costs. | |
HN Discussion:
| |
| 18. | In Emacs, everything looks like a service(yummymelon.com) |
| 244 points by kickingvegas 1 day ago | 102 comments | permalink | |
tl;dr: Emacs isn't an OS, but its built-in libraries for UI (minibuffers, completion), networking (URL, sockets, JSON/XML parsing), and data storage (hash tables, SQLite) make it trivial to build client applications for any service via Elisp. The post demonstrates this with a 67-line wttr.in weather client that fetches and parses JSON, plus an even shorter version that just shells out to a Python script—illustrating how any command-line utility or web API can be treated as a service consumable from within Emacs. | |
HN Discussion:
| |
| 19. | Train sim created by just one person is being called the best ever made(kotaku.com) |
| 828 points by oumua_don17 6 days ago | 341 comments | permalink | |
tl;dr: Running Train, an Early Access train sim built by solo developer Novatetsu Games, is drawing rave Steam reviews for its meticulously detailed fictional Japanese setting—40km of track with logically placed powerlines, traffic, temples, and dynamic weather/seasons. It offers 42 routes across two fake rail lines, supports the Zuiki MASCON peripheral, and can be played hands-on or set to auto-drive while you explore via free camera. It's $18, with plans to expand to 100km of track and add passengers and a conductor mode by end of next year. | |
HN Discussion:
| |
| 20. | John Deere owners will get the right to repair equipment under FTC settlement(apnews.com) |
| 1376 points by djoldman 2 days ago | 300 comments | permalink | |
tl;dr: John Deere has settled with the FTC and five state attorneys general, agreeing to provide diagnostic and repair tools to equipment owners and independent shops rather than restricting them to authorized dealers. The company will pay $1 million to the states, face 10 years of compliance oversight, and is barred from retaliating against those who bypass its dealer network. This follows a separate $99 million class-action settlement Deere reached with farmers in April. | |
HN Discussion:
| |