Apple sues OpenAI, accuses ex-employees of stealing trade secrets

492 stock_toaster 2026-07-10 20:47:09 UTC 9to5mac.com Qwen Share 239

Apple sues OpenAI for trade secret theft by ex-employees, including Tang Tan and Chang Liu, who allegedly stole confidential data and used it in hiring. The lawsuit highlights over 400 Apple alumni at OpenAI, led by Jony Ive, and claims OpenAI exploited Apple's proprietary info for hardware projects.

QuadRF can spot drones and see WiFi through my wall

448 speckx 2026-07-10 15:59:53 UTC jeffgeerling.com Qwen Share 172

The QuadRF is a Raspberry Pi-powered phased-array radio enabling advanced signal processing, WiFi analysis, and RF visualization, with potential for Moon-scale radio experiments. Despite a rough UI, its MIPI-based high-speed data transfer and modular design showcase impressive performance for SDR applications.

New York City to to ban deceptive subscription practices

408 randycupertino 2026-07-10 18:26:24 UTC theguardian.com Qwen Share 210

New York City's Mamdani administration bans deceptive subscriptions and junk fees, enforcing fines for hard-to-cancel services and requiring upfront pricing for all charges, targeting housing and consumer markets. The rules aim to combat hidden costs and corporate malpractice, facing industry opposition but backed by consumer advocates.

Good Tools Are Invisible

358 theanonymousone 2026-07-10 10:32:41 UTC gingerbill.org Qwen Share 166

A good tool should be invisible, not celebrated for complexity or puzzles. Avoid reframing flaws as fun or identity; prioritize productivity over tribalism and honest evaluation.

GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture [pdf]

343 scrlk 2026-07-10 18:29:19 UTC cdn.openai.com Qwen Share 278

The paper proves the Cycle Double Cover Conjecture, asserting every bridgeless graph has a cycle cover where each edge is in exactly two cycles. The proof leverages cubic graphs, Γ-flows, and linear algebra to construct the cover, extending prior results on 8-flows and 3-edge-colorings.

Write code like a human will maintain it

325 ScottWRobinson 2026-07-10 13:33:00 UTC unstack.io Qwen Share 270

Relying on LLMs to generate repetitive code risks perpetuating poor practices, as duplicated logic trains the model to mimic your codebase's patterns. Over time, this creates a self-reinforcing cycle where shortcuts become entrenched, undermining maintainability and requiring manual intervention to correct.

Late Bronze Age Collapse

318 dmonay 2026-07-10 11:59:55 UTC acoup.blog Qwen Share 225

The Late Bronze Age Collapse (c. 1220–1170 BC) saw interconnected states in the Eastern Mediterranean and Middle East decline or collapse, marked by site destructions and political fragmentation. Causes remain debated but likely include climate shifts, warfare, and economic strain, with uneven regional impacts.

AI-generated videos to maximally drive a target brain region

266 smusamashah 2026-07-10 07:39:11 UTC nevo-project.epfl.ch Qwen Share 223

NEvo uses a digital twin and genetic algorithm to evolve videos maximizing brain region activation. Dynamic videos outperform static ones, highlighting regions' preference for motion and social features.

EU Commission: addictive design Instagram and Facebook in breach of the DSA

247 jeroenhd 2026-07-10 11:00:15 UTC ec.europa.eu Llama Share 168

Servers are temporarily unavailable, please try again later. We apologize for the inconvenience caused.

In Emacs, everything looks like a service

230 kickingvegas 2026-07-10 08:21:10 UTC yummymelon.com Qwen Share 99

Emacs functions as an OS-like environment by orchestrating applications and services above the kernel, enabling users to manage computing tasks within its ecosystem. Through Elisp and libraries, it treats external utilities as services, allowing seamless client-server interactions and improvising workflows.

Successful Companies Go Blind

188 speckx 2026-07-10 13:31:08 UTC ianreppel.org Qwen Share 67

Mexican cavefish retain eye genes despite blindness, mirroring companies that lose competence as stable environments suppress innovation, prioritizing comfort over skill. Surviving in predictable conditions, they adapt to internal pressures, mistaking stagnation for loyalty until external change reactivates dormant traits.

Mayor Mamdani announces "Click-to-Cancel" rules

178 thisislife2 2026-07-10 19:30:12 UTC nyc.gov Qwen Share 2

New York City introduces rules to ban hidden junk fees and simplify subscription cancellations, saving residents up to $162.5M annually. The measures, part of Mayor Mamdani’s affordability agenda, require transparent pricing and one-click cancellation, marking a first-in-the-nation consumer protection effort.

How the terrorist group Boko Haram uses frontier AI

176 imustachyou 2026-07-10 18:49:26 UTC casp.ac Qwen Share 148

Terrorists like Boko Haram use AI tools (ChatGPT, Claude, etc.) for attack planning, explosives, and operations via specialized units and transnational training, aided by Islamic State. This systematic AI adoption, surpassing prior estimates, highlights urgent policy and security concerns.

Parental device use and the adolescent-caregiver attachment bond

167 hbcondo714 2026-07-10 00:20:28 UTC frontiersin.org Qwen Share 150

This study validated the Device Attachment Interference Scale (DAIS) to assess adolescents' perceptions of caregivers' device-centric behaviors, finding higher DAIS scores correlated with greater insecure attachment (anxious/avoidant) to both mother- and father-like figures. Results highlight parental tech use as a relational risk factor for adolescent attachment insecurity, emphasizing ...

The tech of 'Terminator 2' – an oral history (2017)

161 markus_zhang 2026-07-10 16:48:41 UTC vfxblog.com Qwen Share 66

In 1991, ILM pioneered CGI tools like 'Make Sticky' and 'Body Sock' to create Terminator 2's liquid metal T-1000, overcoming limited resources and nascent tech. Artists and developers collaborated intensively, blending innovation with practical problem-solving, cementing the film's legacy as a VFX milestone.

Snails' teeth beats spider silk as nature's strongest material (2015)

154 simonebrunozzi 2026-07-10 16:37:52 UTC smithsonianmag.com Qwen Share 125

Marine snails' teeth, composed of goethite nanofibers and protein, are Earth's strongest natural material, exceeding spider silk and Kevlar. Though outmatched by graphene, they rival top carbon fibers, while rare materials like lonsdaleite surpass diamond's hardness.

Punk, or why I don't stream anymore

139 surprisetalk 2026-07-10 13:30:19 UTC geohot.github.io Qwen Share 176

The user critiques the erosion of authentic engagement in hacker culture and digital identity, replaced by performative spectacle and AI-driven inauthenticity. They warn of a homogenized, corporate-dominated internet that commodifies culture, demanding inner reality in a soulless, information-war-ravaged world.

GPT-5.6, Grok 4.5, Claude, and Muse Spark build the same 4 apps

134 hershyb_ 2026-07-10 20:52:28 UTC tryai.dev Qwen Share 79

A model comparison tested 12 AI systems (GPT-5.6 tiers, Meta’s Muse Spark, open-weights like Qwen) across four tasks with five attempts each, revealing performance gaps between leading models and open-source alternatives, while raw results are publicly accessible for self-evaluation.

A love letter to flashcards

123 surprisetalk 2026-07-10 15:30:44 UTC lesleylai.info Qwen Share 79

Initially dismissing flashcards for deep learning, the author adopted spaced repetition via Anki, creating personalized cards for concepts and insights rather than rote facts. This approach enhanced long-term retention and enabled resuming learning after breaks.

War Atlas: An interactive cartography of every named war in human history

111 NaOH 2026-07-10 17:52:48 UTC waratlas.org Full Share 58

An interactive cartography of 10,584 recorded military conflicts across 5,000 years of human history, with 427 historical empire borders, casualty estimate ranges, and era-by-era timeline navigation.

Please don't discontinue Gemini 2.5 Flash

108 NickDob 2026-07-10 20:00:28 UTC discuss.ai.google.dev Qwen Share 74

User thanks Gemini for models but notes Gemini 3 Flash underperforms 2.5 Flash, critical for Australia's low-latency voice agents (300-400ms). Gemini 3.5 Flash's 700-800ms latency and no local deployment break this use case.

Apple sues OpenAI, accusing it of stealing company secrets

107 jbegley 2026-07-10 21:20:29 UTC nytimes.com Full Share 13

Please enable JS and disable any ad blocker

Focus

99 iacguy 2026-07-10 00:40:33 UTC boz.com Qwen Share 54

The author reflects on Facebook's intense startup era, marked by extreme workloads and relentless focus, contrasting it with later dilution of priorities as the company grew. They argue that even well-intentioned diversions from core goals—like supporting nonprofits—accumulate hidden costs, eroding efficiency and clarity over time.

Europe's Largest Unions Demand Right to Cancel Work on Days Above 30C

88 robtherobber 2026-07-10 12:59:49 UTC novaramedia.com Qwen Share 125

European unions demand EU heat protection laws, proposing 30°C/32.5°C thresholds for jobs to prevent 277k injuries and 230 deaths annually. UK TUC also calls for max temps, with 64k+ petition signatures, urging legal cooling and climate action.

Computation as a universal and fundamental concept

83 simonpure 2026-07-10 15:23:42 UTC ergo.org Qwen Share 70

Tim Roughgarden explores computational limits, starting with Turing's 1936 proof that some problems (like the halting problem) are unsolvable by algorithms. He then examines NP-completeness, showing that many problems resist efficient solutions, leading to the unresolved P vs NP question with implications for cryptography and AI.

45% of Enthusiasts 'Seriously Considering' Leaving Sony for PC

80 speckx 2026-07-10 18:41:10 UTC pushsquare.com Qwen Share 108

A Push Square poll reveals 45% of PlayStation fans consider switching to PC gaming, citing Sony's shift to digital-only games and PS Store flaws. Concerns over PS6's $1,000 price and lack of store improvements drive 29% to opt for PCs or delay purchases.

Einstein's relativity rules chemical bonds in heavy elements, new research shows

79 hhs 2026-07-10 22:30:44 UTC brown.edu Qwen Share 36

Brown University researchers discovered relativistic effects in heavy elements like bismuth blur sigma/pi bonds in triple bonds, challenging textbook models. Photoelectron spectroscopy revealed hybridized bonds, impacting chemistry education and applications in solar cells and quantum tech.

Ben Bernanke Joins Anthropic Oversight Trust

78 Jimmc414 2026-07-10 02:39:07 UTC anthropic.com Qwen Share 81

Anthropic's LTBT appointed Dr. Ben Bernanke, a Nobel-winning economist and former Fed Chair, to guide responsible AI development. The independent trust advises on AI's economic impacts and ensures long-term societal benefits.

After 7 years in production, Scarf has reluctantly moved away from Haskell

77 aviaviavi 2026-07-10 13:30:41 UTC avi.press Qwen Share 84

A long-time Haskell advocate, the author moved Scarf's new development to Python due to AI-driven workflows and Haskell's compilation/ecosystem bottlenecks, urging the community to prioritize AI-era optimizations like faster builds, better tooling, and agent-friendly design to remain relevant.

An update on residential proxies and the scraper situation

76 chmaynard 2026-07-10 19:38:34 UTC lwn.net Qwen Share 65

Websites are overwhelmed by scraper bots using residential proxies and compromised devices, evading detection through fake user behavior and coordinated attacks. Defensive measures like traffic analysis and temporary takedowns offer limited relief as attackers adapt, underscoring the need for ethical AI training data solutions.

Unified Memory, Explained: Why Mini PCs Can Run 70B Models a Big GPU Can't

70 ermantrout 2026-07-10 10:40:26 UTC vettedconsumer.com Qwen Share 55

Mini PCs with unified memory (e.g., AMD Strix Halo) can host large 70B LLMs due to 128GB shared memory, but generate text slowly (4-6 tok/s) due to low bandwidth (256GB/s), unlike GPUs (e.g., RTX 5090) with high bandwidth (1,792GB/s) but limited 32GB VRAM. Their strength is capacity, not speed, excelling when models exceed GPU VRAM limits despite slower inference.

SpaceX wants to launch 100k more Starlink satellites for 100x the bandwidth

67 CrankyBear 2026-07-10 17:51:07 UTC zdnet.com Qwen Share 211

SpaceX seeks FCC approval for 100,000 Gen3 Starlink satellites to deliver ultra-low-latency, multi-gigabit broadband, requiring hardware upgrades and facing interference risks. The plan aims to expand satellite internet access but encounters regulatory challenges, competition, and astronomical objections.

Build your own vulnerability harness

63 ianrahman 2026-07-10 01:28:43 UTC blog.cloudflare.com Qwen Share 23

The blog outlines a model-agnostic security architecture using interchangeable AI models in a multi-stage pipeline (Recon, Hunt, Validate) for enterprise codebases, enabling cross-repo vulnerability detection and reducing false positives through adversarial validation. It emphasizes orchestration over single agents, leveraging deterministic code and LLMs to trace dependencies, generate proofs ...

ActivityPub over ATProto (2023)

57 albuic 2026-07-10 13:40:01 UTC berjon.com Qwen Share 38

The article critiques social media's divisive "factional" dynamics, proposing integration of AT Protocol and ActivityPub to break silos. It argues combining their strengths—user agency and web-native design—could create a more open, interoperable social web.

Ryanair passenger sucked toward broken window after midair engine failure

56 amelius 2026-07-10 12:31:50 UTC simpleflying.com Qwen Share 62

A Ryanair flight from Greece to Germany returned after an engine failure shattered a window, injuring a passenger partially sucked out. The crew safely landed the plane, arranged a replacement flight, and authorities are investigating, similar to past incidents like Southwest Airlines' 2018 tragedy.

Java 27: What's New?

51 loicmathieu 2026-07-10 10:08:49 UTC loicmathieu.fr Qwen Share 54

Java 27 adopts G1 as the universal default GC, enhances TLS 1.3 with quantum-resistant hybrid key exchanges, enables compact object headers by default, and masks sensitive data in JFR recordings. It previews Valhalla-related features like lazy constants and primitive pattern matching, while removing JVMCI and preparing for future Valhalla integration in Java 28.

The Annotated JEPA

50 surprisetalk 2026-07-10 13:45:05 UTC elonlit.com Qwen Share 10

This post provides a step-by-step guide to implementing Joint Embedding Predictive Architectures (JEPAs), Yann LeCun's self-supervised learning framework that predicts latent representations instead of pixels, focusing on image (I-JEPA) and video (V-JEPA) models. It emphasizes using exponential moving average (EMA) for stable training, structured masking strategies, and predictor networks to ...

Anyone else get a vague GitHub shakedown notice?

48 jjgreen 2026-07-10 16:57:35 UTC lists.osgeo.org Qwen Share 42

Greg received a GitHub billing email for Code Quality despite no paid plan, having left orgs, and no sponsorship, questioning if QGIS activity triggered it and if others faced similar unexpected charges. Concerns arise over potential billing errors or scams.

Cpp2Rust: Translates C++ to safe Rust automatically

44 signa11 2026-07-10 16:24:20 UTC github.com/cpp2rust Qwen Share 11

Cpp2Rust automatically translates C++ to safe Rust using clang's AST, generating code with libcc2rs for pointer semantics. It offers an unsafe mode via --model=unsafe and employs Ptr<T> to manage C pointers safely under Rust's borrow checker.

Show HN: Runloom – Go-style coroutines for Python free-threaded

43 Uptrenda 2026-07-10 13:07:38 UTC github.com/robertsdotpm Qwen Share 22

runloom enables Go-style stackful coroutines in Python, using a work-stealing scheduler and netpoll for I/O, allowing millions of coroutines across cores in free-threaded CPython 3.14t (GIL off). It supports blocking code via monkey patching and matches Go's concurrency performance, though with higher memory per coroutine.

Software Engineer's Firing Ruled Illegal in a Rare Win for a Tech Worker

41 ytpete 2026-07-10 20:29:55 UTC nytimes.com Full Share 24

Please enable JS and disable any ad blocker

Hands-On with the AMD Ryzen AI Halo

41 bdcravens 2026-07-10 16:06:38 UTC microcenter.com Qwen Share 40

The AMD Ryzen AI Halo is a compact desktop with 128GB unified memory and a Ryzen AI Max+ 395 chip, enabling local AI model execution to cut cloud costs. It efficiently runs large LLMs and image generation tasks, offering a server-like setup for offline AI workflows.

You're Not a Better Engineer Because You Type Git Commands by Hand

39 meerita 2026-07-10 17:01:25 UTC minid.net Qwen Share 126

After 27 years, I delegate repetitive engineering tasks like Git hygiene, PRs, and code comments to AI, freeing time for core work. AI ensures consistency, reduces errors, and transforms repositories into coherent systems, making manual "craft" obsolete and cost-inefficient.

Star Just Ate a Planet, and It's Not Done Yet

38 wglb 2026-07-10 01:20:16 UTC nytimes.com Full Share 55

Please enable JS and disable any ad blocker

Proton AG Services is currently experiencing some issues

35 exploraz 2026-07-10 13:36:56 UTC status.proton.me Full Share 37

Proton Services's Status Page - Investigating Incident.

Ask HN: How do you use Vim in the era of AI?

34 rstagi 2026-07-10 13:11:30 UTC news.ycombinator.com Qwen Share 73

The user leverages Vim (neovim) with tmux and LSP for code navigation/editing, using AI as a companion tool for prompt refinement and inline suggestions rather than full code generation. They prioritize a terminal-based workflow, integrating Vim plugins and CLI for seamless LLM interaction and consistent remote/local environments.

GhostLock, a stack-UAF that has existed in ALL Linux distributions for 15 years

33 djfergus 2026-07-10 20:43:02 UTC nebusec.ai Qwen Share 8

GhostLock (CVE-2026-43499) is a Linux kernel use-after-free vulnerability in rtmutex code, enabling privilege escalation via a dangling pointer from a freed stack frame. Exploited through a race condition and controlled memory overwrite, it allows unprivileged users to execute arbitrary code, fixed in Linux 7.1.

Materials innovation has a scale-up problem, not discovery

24 groznyj 2026-07-10 19:03:36 UTC atomscale.ai Qwen Share 10

Richard Feynman's 1959 vision of atomic-scale engineering enabled modern tech, but scaling materials remains a bottleneck. Atomscale uses AI to transform fragmented data into actionable insights, accelerating material production from lab to real-world applications.

Ask HN: What was the last task where only a frontier model could do it?

22 thedebuglife 2026-07-10 18:02:02 UTC news.ycombinator.com Qwen Share 26

Frontier models like Fable and GPT outperform open models (GLM/Qwen) in complex tasks like code audits, tax problem-solving, and multi-service repo analysis, offering higher accuracy and depth. While open models suffice for simpler tasks, frontier models excel in sustained agent work, security audits, and nuanced problem-solving, justifying their cost for critical applications.

The first programmer I know who was murdered in a targeted strike was Haitham

21 lr0 2026-07-10 19:18:24 UTC lrb.co.uk Qwen Share 3

The text highlights Gaza's humanitarian crisis from Israel's blockade and military strikes, displacing families and destroying livelihoods, while critiquing failed non-violent resistance and Hamas's strategies. It calls for Palestinian unity, democratic reforms, and international accountability to end the cycle of violence and neglect.

AI doesn't know how to forgive and cannot forget

21 tejassuds 2026-07-10 12:37:40 UTC tejassuds.com Qwen Share 23

AI systems retain data perfectly without decay, lacking human-like forgetting and forgiveness, which involve active memory pruning and intentional release. This creates challenges in AI-human interactions, as forgiveness requires intentional release, not just data deletion.

Offline tool to stabilize TV volume using IR control and spike detection

19 Bo_Amigo_910 2026-07-10 13:39:49 UTC github.com/adbusterofficial Llama Share 9

AdBuster 2.0 PRO is a Windows tool that automatically lowers loud TV ads and restores normal volume smoothly. It uses real-time audio analysis, machine learning, and Broadlink IR control for intelligent volume control.

Man nearly sucked out of 'detached' window on Ryanair flight

16 theanonymousone 2026-07-10 13:40:31 UTC rte.ie Qwen Share 4

A Ryanair flight window detached mid-air over North Macedonia, nearly pulling a Serbian tourist out; he sustained burns but survived. Ryanair diverted the flight back to Thessaloniki and provided a replacement aircraft.

Browser Fingerprinting – How websites track you across internet –without cookies

16 hackstar 2026-07-10 07:07:50 UTC mysysinfo.com Qwen Share 5

Browser fingerprinting tracks users by collecting device/browser data (e.g., OS, fonts, GPU) without cookies, working in incognito and persisting across sessions. It's used for ads, security, and fraud prevention, countered by browsers like Brave or Firefox with privacy settings.

Show HN: SubjectiveZero, an open-source agentic node editor for creative coding

13 tasoeur 2026-07-10 15:23:50 UTC sxp.studio Qwen Share 4

SubjectiveZero uses AI agents to generate editable code nodes for real-time visual effects, integrating with tools like Claude and Codex. It offers collaborative editing, live rendering, and open-source AGPL-3.0 licensing with no black boxes.

Garnix Is Joining Shopify

12 l2dy 2026-07-10 16:21:18 UTC garnix.io Full Share 10

Sadly, as part of this transition, the hosted garnix service will shut down on July 15th 2026. But we are open sourcing the garnix codebase, available here; we hope this will help you smoothly move to using your own instance or a shared one, and maybe help the project live on and evolve. If you are interested in operating a public community instance, please get in touch — we’d be happy to talk.

Show HN: I built a free app for New Yorkers to save money on groceries

12 eli_yumnik 2026-07-10 00:17:36 UTC sbnyc.app Full Share 10

Compare this week's grocery deals across 690+ NYC stores in all five boroughs. Weekly-ad prices refreshed every morning, live store prices, cashback — free, no login.

Prismata: Confining cross-site prompt injection in web agents

11 zhinit 2026-07-10 21:03:38 UTC arxiv.org Qwen Share 0

Autonomous web agents inherit XSS-like risks via prompt injection. Prismata mitigates this by dynamically assigning trust-based permissions and confining access without developer annotations, reducing attacks while preserving task utility.