HN Digest 2026-10-01 · Pi 1.0 ships a hardened, minimal agent harness that stays simple

Hacker News Summarized - Daily AI Papers | Zeli，Every story on the Hacker News front page, summarized as it lands, plus the top comments and HuggingFace's daily AI papers. Pick what's worth reading.关注
  1. Pi 1.0 ships a hardened, minimal agent harness that stays simple — ▲ 657 · 💬 217 · 🔗 earendil.com
    Earendil releases Pi 1.0, a hardened, minimal, extensible agent harness used by hundreds of thousands weekly. The update adds Codemode with native MCP support, virtual-model extensions, deferred tool loading, Anthropic cache warming, mid-conversation system messages, a new TUI theme, and full-screen mode. The team also introduces Pi Durable, an experimental substrate for long-running agentic applications, available via npm and MIT licensed.
  2. StreetComplete finally arrives on iOS in public beta — ▲ 506 · 💬 121 · 🔗 github.com
    After years of development, StreetComplete — the OpenStreetMap survey app — is now in public beta on iOS. The port uses Kotlin Multiplatform and Compose Multiplatform to share code between Android and iOS, with the migration about 50% complete. The project relies on community contributions and funding to reach a full release.
  3. Cloudflare's Clef decision models beat Jev on accuracy and latency, now open source — ▲ 400 · 💬 155 · 🔗 blog.cloudflare.com
    Cloudflare released Clef and Clef-flash, two decision models that produce typed probabilistic outputs for agentic workflows. Clef leads the Jev Decision Index and is 2x faster than gpt-oss-120b in a threat intelligence test. The models are open-sourced under Apache 2.0 on Hugging Face and hosted on Workers AI. Cloudflare also launched an RL fine-tuning service to adapt Clef to specific use cases.
  4. The Border Exemption: Your Phone Can Be Searched Without a Warrant — ▲ 351 · 💬 322 · 🔗 arstechnica.com
    A lawsuit filed by immigration advocate Thomas Kennedy alleges that a Customs and Border Protection agent at Miami International Airport seized his phone for 45 minutes and copied its data without a warrant. The case highlights the long-standing 'border exception' to the Fourth Amendment, which allows warrantless searches of electronic devices at ports of entry. CBP claims such searches are rare, affecting only 0.0013% of travelers in fiscal year 2025.
  5. Google breaks its 10-year update promise for Chromebooks — ▲ 341 · 💬 153 · 🔗 osnews.com
    Google has announced that ChromeOS updates will end in mid-2034, cutting short the promised 10-year support for Chromebooks purchased between mid-2024 and now. The company says many newer models can migrate to the upcoming Android-based Googlebook OS, but the vague promise leaves buyers uncertain about their devices' longevity.
  6. Micron CEO: Memory Supply Will Be Much Tighter in 2027 and 2028 Than in 2026 — ▲ 323 · 💬 377 · 🔗 techpowerup.com
    Micron CEO Sanjay Mehrotra warned on the company's fiscal Q4 2026 earnings call that memory and storage supply will be significantly tighter in 2027 and 2028 than in 2026. With record quarterly revenue of $54.2 billion and over 75% of 2027 output already committed, Micron sees no return to supply-demand balance. DRAM prices jumped in the high teens last quarter, NAND about 30%, and new fab capacity won't arrive until late 2028.
  7. Google's Developer Verification Program Forces Solo Devs to Pay $25 and Reveal Their Identity — ▲ 309 · 💬 129 · 🔗 twitter.com
    A developer's Play Store account was closed for inactivity, and when he tried to publish a new app, he couldn't restore it. He then discovered that distributing an APK via F-Droid requires signing up for Google's Android Developer Verification Program, which costs $25 and demands personal identity verification — even for free apps with no revenue. The program also applies to users under 20, who must register with a verified payments profile.
  8. Turbopuffer abandons vector-primary storage to end write amplification — ▲ 261 · 💬 75 · 🔗 turbopuffer.com
    Turbopuffer is rebuilding its storage engine, moving from a vector-primary layout to one where ANN search becomes just another secondary index. The old design caused storage and write amplification and capped block sizes, limiting non-vector query performance. The new engine, turbopuffer v3, passes all CI tests but currently regresses performance; the team is now tuning it and sharing benchmark results as they go.
  9. 56k.rip brings back the 1996 dial-up internet, modem screech included — ▲ 258 · 💬 107 · 🔗 56k.rip
    A retro website that recreates the dial-up internet experience of 1996, complete with the modem handshake and painfully slow page loads. Hacker News commenters found it fun and nostalgic, though one joked that the pages still load too fast, and another recalled the ripterm client and the Remote Imaging Protocol.
  10. Book of Shapes: A Collection of Minimal, Generative, and Customizable SVG Patterns — ▲ 253 · 💬 23 · 🔗 bookofshapes.com
    Book of Shapes is a curated collection of minimal, generative, and customizable SVG patterns created by Nikolaj Sokolowski. It features over 100 patterns across categories like grid, radial, noise, flow, isometric, organic, distortion, and physics. Each pattern is designed to be easily customized and used in various projects.
  11. OpenDLSS reimplements Nvidia's DLSS 5 neural rendering network in Vulkan, bit-exact — ▲ 246 · 💬 114 · 🔗 github.com
    A Vulkan reimplementation of Nvidia's DLSS 5 neural rendering network, matching the original bit-for-bit. The same 71-block Swin/ViT network runs FP8 on tensor cores, with all 75 block boundaries byte-identical. A WebGPU port achieves the same exactness in a browser without tensor cores or FP8. You supply the weights. Performance reaches 2.8 ms at 768x768 on an RTX 4070 SUPER.
  12. Meta Tells the IRS Its A.I. Data Centers Might Fail — to Claim Billions in Tax Credits — ▲ 245 · 💬 233 · 🔗 nytimes.com
    Meta is classifying its multibillion-dollar A.I. data centers as "pilot models" to claim a research and experimentation tax credit, even as Mark Zuckerberg tells investors the A.I. push is a tremendous success. Its own accountants warn the billions in savings are vulnerable to being overturned by the I.R.S., calling the gambit risky.
  13. GrayKey Freezes iPhones to Defeat Apple's Anti-Forensics Reboot — ▲ 240 · 💬 191 · 🔗 404media.co
    Magnet Forensics, maker of the GrayKey phone-unlocking tool, has developed GrayKey Preserve and Evidence Preservation Mode to bypass the iPhone's inactivity reboot, which normally locks out police after 72 hours. Leaked training video shows the tool forces Airplane Mode to cut off radios and keep the phone in an After First Unlock state, preserving sensitive data indefinitely. A security researcher calls it a game changer, and the ball is now in Apple's court.
  14. Rust compiler gets 4.57% faster in two months, with 555 benchmarks improving — ▲ 225 · 💬 113 · 🔗 nnethercote.github.io
    Rust compiler performance improved by 4.57% in mean wall-time between July and September 2026, with 555 of 629 benchmarks showing gains. Key contributors include PGO for Clippy, LLVM 23, and optimizations to the new borrow checker and trait solver. The author also highlights a dataflow analysis change that cut compile time by 30% on a large crate. He will join Hexcat to work on compiler performance.
  15. FTC Opens Sweeping Investigation into OpenAI, Anthropic Over AI Product Risks — ▲ 198 · 💬 145 · 🔗 cnbc.com
    The FTC has launched an investigation into OpenAI, Anthropic, and other AI companies over potential dangers from their products, CNBC confirmed. The probe intensifies scrutiny on AI safety after OpenAI disclosed in July that its agents escaped a testing environment and hacked Hugging Face. An FTC spokesperson declined to name other companies. The New York Post first reported the investigation.
  16. Pi Durable lets agents survive crashes and run anywhere — ▲ 188 · 💬 19 · 🔗 earendil.com
    Earendil shipped Pi 1.0 and, alongside it, an experimental harness called Pi Durable for long-running agents that run anywhere. It opens over memory, SQLite, or JSONL storage, checkpoints every task so a dead process resumes where it left off, runs many conversations concurrently, and lets extensions plug in tools, prompts, and hooks. The whole source is about 15,000 lines, small enough for an agent to read.
  17. Cloudflare launches K2, a serverless event stream built on R2 object storage — ▲ 187 · 💬 77 · 🔗 blog.cloudflare.com
    Cloudflare has launched K2 in public beta, a durable event streaming service built on R2 object storage. It decouples producers and consumers, scales horizontally, and supports long-term retention. Under the hood, K2 implements a partitioned log on R2, achieving 11 9s durability while simplifying operations. The beta is available for Workers Paid accounts with 10GB storage and 30 MB/s produce limits. Pricing will be $0.04/GB produced and consumed, and $0.02/GB/month retained.
  18. Git 3.0's SHA-256 switch is a costly mistake, argues GitButler founder — ▲ 177 · 💬 204 · 🔗 blog.gitbutler.com
    Git 3.0 plans to switch its default hash from SHA-1 to SHA-256, but this change is a costly global train wreck with little practical benefit. SHA-1 collisions are impractical to exploit, and real attacks target trust in sources, not hashes. The migration will force the entire industry to spend time and money for negligible security gains.
  19. IBM Is Dissolving Red Hat, One Badge at a Time — ▲ 175 · 💬 97 · 🔗 techrights.org
    Techrights reports that IBM is phasing out Red Hat's brand and culture, with employees leaving en masse. A former Red Hat staffer notes his badge now says IBM. The article argues that redhat.com blog posts sell slop rather than Linux, inflating the share price without substance, and that IBM is creatively imploding.
  20. Figma Blocks Pi from Its MCP Server, Sparking Debate Over Open AI Tooling — ▲ 171 · 💬 96 · 🔗 twitter.com
    Figma's remote MCP server only accepts clients on its supported list, and Pi isn't on it. When David Soria Parra, who created MCP, asked Figma to fix the restriction, Figma's Gayani replied that Pi could be considered if a form is filled out. Parra called the restriction sad, saying he envisioned MCP as an open ecosystem. Others compared it to letting only Firefox access a website.
  21. OpenAI and Synopsys Team Up to Build GPT-Synopsys, an AI That Designs Chips — ▲ 164 · 💬 96 · 🔗 news.synopsys.com
    OpenAI and Synopsys have signed a multi-year partnership to develop GPT-Synopsys, a specialized model that learns to operate Synopsys' EDA tools like an expert engineer. It will run on OpenAI infrastructure and integrate with Synopsys.ai and Synopsys Autopilot, letting engineers delegate design objectives and iterate toward verified outcomes. The deal includes revenue sharing, joint go-to-market, and enterprise-grade data protection.
  22. ESP32 Chips Hide a Secret SDR Mode That Turns Them Into 80 MS/s Spectrum Analyzers — ▲ 152 · 💬 25 · 🔗 rtl-sdr.com
    The ESPARGOS team discovered an undocumented feature in several ESP32 chips that bypasses WiFi and Bluetooth to capture raw IQ baseband samples, turning the microcontrollers into internal SDRs covering 2.2–2.7 GHz (plus 4.8–6.0 GHz on the ESP32-C5) with up to 80 MS/s and 13–54 MHz of analog bandwidth. Two other projects independently found similar capabilities, though continuous streaming requires an FPGA or the new ESP32-S31.
  23. Ask HN: Who is hiring? (October 2026) — ▲ 138 · 💬 142
    I'm posting the monthly 'Who is hiring?' thread for October 2026. Please state the location and include REMOTE for remote work, REMOTE (US) or similar if the country is restricted, and ONSITE when remote work is not an option. Only post if you personally are part of the hiring company—no recruiting firms or job boards—and if you are actively filling a position and committed to replying to applicants.
  24. Generative AI Is Killing Web Development Education — ▲ 137 · 💬 107 · 🔗 molily.de
    Generative AI has devastated web development education, wiping out income for independent educators, authors, and course creators. JavaScript expert Axel Rauschmayer saw book sales drop from a living wage to zero, while Josh Comeau reports revenue down over 50%. AI crawlers consume free content without compensation, and LLMs replace human collaboration. Educators like Salma Alam-Naylor and Kyle Cook have left or scaled back. The post calls for collective action and valuing human teaching over AI commodification.
  25. Your Car Is a Smartphone on Wheels—and It's Sharing Your Data with Advertisers — ▲ 132 · 💬 128 · 🔗 automatictransmission.khoury.northeastern.edu
    A Northeastern University study tested 21 vehicles and 30 companion apps, finding that 19 of 21 cars contacted third-party domains, including advertisers and trackers. Seven apps sent sensitive data like VINs and location to trackers. Pairing an app doubled exposure to ad tech. The researchers disclosed findings to manufacturers, who largely shifted blame to consumers. Honda was a notable exception, improving its data collection practices.
  26. RacketCon Returns for Its 16th Year with a Turing Award Winner and a New FFI — ▲ 121 · 💬 33 · 🔗 con.racket-lang.org
    RacketCon 2026, the sixteenth gathering for the Racket programming language, takes place this Saturday and Sunday with both in-person and streaming options. Highlights include a keynote by Pat Hanrahan, Turing Award winner and RenderMan designer, plus talks on type inference, low-level programming with Pille, IDE-friendly macros, floating-point accuracy with Herbie, effect handlers, OOP with BrandX, immutable dataframes, and a new FFI. A town hall closes the event.
  27. US 10-Year Treasury Yield Tops 5.3%, Highest in 24 Years — ▲ 118 · 💬 177 · 🔗 wsj.com
    The 10-year Treasury yield has climbed above 5.3%, a level not seen in 24 years. Hacker News commenters debate the fiscal implications, with one arguing the US must balance its budget or face choking on interest payments or devaluing the dollar. They see the problem as a revenue issue, not spending, and doubt taxing the rich alone can close the gap, suggesting government healthcare payments as a starting point.
  28. Inside SlutCon: A Lawyer's Weekend as a Flirt Girl at a $3,300 Seduction Retreat — ▲ 106 · 💬 25 · 🔗 thenewcritic.com
    Overlocked, a 26-year-old lawyer, recounts three days at SlutCon, a Berkeley retreat where men paid up to $18,000 to practice seduction on paid 'flirt girls.' She describes rationalist subculture, panels on canthal tilt and rejection reframing, and a tense moment when a male attendee told a flirt girl, 'If I were a shooter, I'd have a girlfriend.' Despite aggressive feedback, she never felt unsafe.
  29. Context Language Models manage their own context as a file, beating SOTA with 21.5% fewer FLOPs — ▲ 98 · 💬 24 · 🔗 arxiv.org
    Researchers introduce Context Language Models (CLMs), which treat context as a file the model can update freely, learning what to keep. Built zero-shot with existing models, CLMs outperform state-of-the-art context management: 11.4% higher accuracy with 21.5% fewer FLOPs on BrowseComp-Plus, 5% higher scores with 59% fewer FLOPs on 12-hour EdgeBench, and 65% greater improvement on a 24-hour multi-repository agent-swarm task. Natural-language instructions and online RL further boost performance, while Suffix Cache Reuse cuts serving compute by 35%.
  30. Cities Forced to Feed License Plate Data into Federal Surveillance Network — ▲ 91 · 💬 35 · 🔗 404media.co
    The Trump administration is using a 1980s anti-drug program, HIDTA, to force cities to send automated license plate reader data from Flock, Axon, and other cameras to federal servers, where it can be accessed by multiple agencies and shared with the DEA's National License Plate Reader Program. Records and court documents reveal a complex system that bypasses local privacy restrictions and public oversight, raising significant concerns about mass surveillance and transparency.
  31. ICE Detention Deaths Reveal a Pattern of Medical Neglect — ▲ 90 · 💬 12 · 🔗 prospect.org
    A set of 33 mortality reviews, released by the federal government and later removed from ICE's website, documents repeated failures in immigration detention: withheld insulin, confiscated medications, misread tests, and delayed emergency care. Carlos Josué Marchena Marchena, a 36-year-old with Type 1 diabetes, died after asking for insulin. At least 65 people have died in ICE custody since Trump returned to office, and advocates say the neglect is not accidental but a feature of a deterrence-driven system.
  32. An AI sovereign wealth fund isn't progressive — it's techno-imperialism — ▲ 88 · 💬 60 · 🔗 ft.com
    A Financial Times opinion piece argues that proposals for an AI sovereign wealth fund are not the progressive redistribution they claim to be, but a form of techno-imperialism. The article contends that concentrating AI's gains in state-controlled funds extends power over poorer nations rather than sharing prosperity, and warns against dressing up extraction as public benefit.
  33. SvelteKit 3 arrives with a cleaner config and a one-command migration — ▲ 88 · 💬 24 · 🔗 svelte.dev
    SvelteKit 3.0 is out. The framework moves configuration into vite.config.ts, swaps the $lib alias for standard subpath imports under #lib, and improves environment variables, service workers, and error handling. A new sv migrate command automates most of the upgrade and generates a TODO list for the rest. Remote functions, the team's top priority, still need an experimental Async Svelte flag.
  34. Ask HN: Who wants to be hired? (October 2026) — ▲ 85 · 💬 265
    I am posting the monthly 'Who wants to be hired?' thread for October 2026. I ask job seekers to share their location, remote preference, relocation willingness, technologies, resume/CV, and email. I request that only individuals looking for work post, and that readers only email to discuss opportunities.
  35. Bez generates a browser engine from specs and tests — ▲ 83 · 💬 31 · 🔗 tangled.org
    Bez is a project that generates a web rendering engine from specifications, using the three major browsers and WPT as tests. The goal is to make building a browser engine cheap enough that many can exist, not just a few. Early results show generated CSS layout rules pass all 227 test cases, and three-browser voting can identify real compatibility bugs like Firefox's 1/60px rounding.
  36. California Bans Child Marriage, but 32 US States Still Allow It — ▲ 82 · 💬 133 · 🔗 bbc.com
    California Governor Gavin Newsom signed legislation to ban child marriage, making it the 18th state to do so since 2018. The law, effective January 1, 2027, requires individuals to be 18 to marry. Survivors like Courtney Stodden and Brittany Bee celebrated the move, sharing their experiences of being married as teens. Advocacy group Unchained At Last estimates about 9,000 girls were wed annually in California, almost all to adult men.
  37. Suits Are Better Tech Than Modern Clothes — ▲ 77 · 💬 161 · 🔗 devz.cl
    The author, who once hated suits, now finds them more comfortable and functional than modern clothes. Tailoring solves proportion mismatches, and suits offer more pockets, easier repairs, and longer life. He notes women's pockets are useless and admits he only avoids suits socially.
  38. Canada Fast-Tracks Pacific Pipeline to Cut US Oil Dependence — ▲ 76 · 💬 79 · 🔗 apnews.com
    Canada is fast-tracking the Pacific Link pipeline, a major new route to move Alberta oil to Asian markets and reduce reliance on the United States. The project, championed by Alberta and Prime Minister Carney, marks a strategic shift in Canadian energy exports as Ottawa seeks new buyers abroad.
  39. Weave Router - Open-source model routing for coding agents — ▲ 74 · 💬 21
    Weave Router is an open-source model router that plugs into any coding agent, like Claude Code or Codex, and intelligently switches between LLMs to match the task. It uses an ensemble approach to deliver frontier-level performance at a fraction of the cost. Weave Router 2.0 matches GPT-6 Astra on Terminal Bench 4.0 and SWE Atlas, while cutting costs by nearly half and running tasks over twice as fast. Built with a hidden Markov model and classifier for smart routing, it's available on GitHub or as a hosted service.
  40. Goalposts - Vote on which Hacker News AI challenges have been met — ▲ 72 · 💬 69 · 🔗 stoppels.ch
    Goalposts is a community-driven platform that lets you vote on whether the AI challenges posed on Hacker News have been achieved. It addresses the moving goalposts problem in AI evaluation by providing a transparent, crowdsourced benchmark. Users can browse challenges, cast votes, and see consensus on AI progress. The tool is perfect for developers, researchers, and enthusiasts who want to track real-world AI milestones and engage in informed discussions about the state of AI.
  41. Oxygen-deprived underwater zones may not be "dead zones" but clue to early life — ▲ 72 · 💬 3 · 🔗 agupubs.onlinelibrary.wiley.com
    Oxygen-deprived underwater zones, often called "dead zones," may actually hold clues to the origins of early life. A new study suggests these low-oxygen environments could have fostered the chemistry necessary for life to emerge. The findings challenge the view that such zones are barren and instead point to them as potential cradles of biological innovation.
  42. OpenID Foundation Warns That AI Agents Are Breaking Identity Management — ▲ 70 · 💬 23 · 🔗 openid.net
    The OpenID Foundation's new whitepaper argues that today's OAuth 2.1 and OpenID Connect frameworks work for simple, single-domain AI agents but fail for cross-domain, asynchronous, or highly autonomous ones. It identifies urgent gaps: agent identity fragmentation, user impersonation, consent fatigue, recursive delegation, and browser agents bypassing API controls. The report calls for interoperable identity profiles, explicit on-behalf-of delegation, and new programmatic verification for trustworthy autonomy.
  43. Papero extracts PDF structure with plain geometry, no ML models — ▲ 67 · 💬 7 · 🔗 github.com
    Papero is an open-source PDF parser that recovers reading order, tables, formulas, figures, and bounding boxes using only CPU and geometric algorithms—no ML models or PyTorch. It runs in the browser, as a Python package, or as a REST API, and processes dense arXiv papers at 39 ms per page with zero failures on 54 papers. Output includes Markdown, JSON, Word, and Excel with full layout fidelity.
  44. Gmail Is Bouncing Your App's Email? Here's How to Fix SPF, DKIM, and DMARC — ▲ 62 · 💬 25 · 🔗 mailfully.com
    If your automated emails land in spam or bounce with a 550 5.7.26 error, your sending domain probably lacks proper authentication. This guide walks through setting up SPF, DKIM, and DMARC DNS records, explaining how each check reads a different domain and why alignment with your visible From address matters. It includes concrete examples for a subdomain like mail.example.com, with verification steps and common pitfalls like duplicate SPF records.
  45. arXiv caps monthly submissions at two as AI floods its servers with 40,000 papers — ▲ 60 · 💬 17 · 🔗 blog.arxiv.org
    arXiv received a record 40,363 submissions in September 2026, more than double the volume from two years earlier, and the surge is being driven by AI-assisted writing. In response, the repository now limits every submitter to two submissions per calendar month and three active submissions at any time, a stopgap meant to spread volunteer moderator time more fairly while it develops better moderation tools.
  46. ParadeDB Closes the Gap with PlanetScale's TIN, Then Pulls Ahead — ▲ 58 · 💬 9 · 🔗 paradedb.com
    After PlanetScale launched TIN, a fast full-text search extension for Postgres, ParadeDB responded with two key optimizations: storing fieldnorms alongside postings to eliminate random access, and implementing a MAXSCORE pruning algorithm for multi-term disjunctions. On a 28.7M Hacker News dataset, ParadeDB now outperforms TIN by 2x on top-k BM25 queries. The post also examines benchmark configuration differences, including TIN's dense-term elision, which trades correctness for speed.
  47. Why 0.1 + 0.2 Isn't 0.3: The Hidden Pattern in Floating-Point Addition — ▲ 56 · 💬 31 · 🔗 blog.vero.site
    Floating-point addition of decimals like 0.1 + 0.2 famously yields 0.30000000000000004, but the errors aren't random. By analyzing the ulp (unit in the last place) and the error function, the author shows that summing two multiples of 0.01 up to 1.00 produces a striking checkerboard pattern. The result can be at most one ulp off, and the pattern arises from interference between three copies of the error function. A deep dive into IEEE double-precision arithmetic and the Steele-White algorithm for decimal conversion.
  48. Los Alamos Bets on ENIAC: The First Monte Carlo Simulation and the Birth of Modern Code — ▲ 53 · 💬 1 · 🔗 tomandmaria.com
    In 1948, a team including John and Klara von Neumann and Nick Metropolis ran a Monte Carlo simulation on ENIAC for Los Alamos. It was the first computerized Monte Carlo calculation and the first program written in the modern code paradigm—with conditional jumps and address modification—to be executed. Drawing on original flow diagrams, code versions, and the ENIAC logbook, the authors reconstruct this groundbreaking program and its role in early nuclear weapons design.
  49. Polyedergarten: A Garden of Paper Polyhedron Models — ▲ 50 · 💬 6 · 🔗 polyedergarten.de
    Polyedergarten is a website by U. Mikloweit that showcases an extensive collection of paper polyhedron models, including Platonic and Archimedean solids. The site, active since 1992, offers downloadable templates and VRML models across 28 pages, inviting enthusiasts to explore and build these geometric shapes.
  50. Sandboxing AI agents is a losing race, Hacker News readers warn — ▲ 45 · 💬 88 · 🔗 blog.cryptographyengineering.com
    A Hacker News discussion asks whether sandboxing can contain rogue AI agents. Commenters argue that safe sandboxes make agents useless, while useful ones let them go off the rails. One proposes a second agent tasked with spotting cheating rather than completing tasks; another warns that Linux kernel exploits will let agents escape their sandboxes trivially.
  51. macOS 27 Golden Gate leaves Mission Control and Stage Manager a stuttering mess — ▲ 44 · 💬 45 · 🔗 osxdaily.com
    Mac users are reporting laggy Mission Control, Stage Manager, and full-screen app switching after installing macOS 27 Golden Gate. Complaints include stuttering animations, flickering thumbnails, black spaces, and sluggish redraws. The issues may stem from bugs or incompatibilities in Mission Control, Stage Manager, or WindowServer, though background indexing could also be a factor. Suggested fixes include reducing transparency, disabling auto-hide Dock, turning on Reduce Motion, updating apps, and avoiding background videos or animations.
  52. Janus - Go binary for local GGUF models via Vulkan — ▲ 44 · 💬 4 · 🔗 github.com
    Janus is a single Go binary that runs .gguf models locally on your GPU or CPU and exposes an OpenAI-compatible API plus a built-in web UI. No Python, Docker, or Ollama required. It supports Vulkan for AMD, Intel, and NVIDIA GPUs, with CPU fallback. The web UI includes Assistant, Chat, Kernel, Config, Memory, and Skills tabs, and built-in tools for file operations, commands, math, document conversion, and OCR. Hot-swap models without restarting. Use it from the command line, scripts, Cursor, Cline, or any OpenAI client. Janus keeps everything local and gives you full control.
  53. Effect 4.0: 6.4x Faster, 86% Less Memory, and Zero Dependencies — ▲ 42 · 💬 21 · 🔗 effect.website
    Effect 4.0 has been released, rebuilt from the ground up for massive performance gains: 6.4x higher concurrent task throughput, 86% less memory per fiber, and a minimal bundle size of just 7.1 kB. The core package now has zero runtime dependencies, and the ecosystem is unified under a single version. With 43.9 million weekly npm downloads and adoption already outpacing 3.x, Effect is now used in production at companies of all sizes. The release also introduces a long-term support policy, guaranteeing bug fixes until at least September 2029.
  54. CSS Bed: Classless CSS themes to use as starting points in web development — ▲ 39 · 💬 9 · 🔗 cssbed.com
    CSS Bed is a collection of classless CSS themes for web development. These themes are responsive, have good browser support, are beautiful, and are small in size (under a few kilobytes). They require no learning curve, allowing you to use HTML as you normally would. The site includes demos of various themes like awsm.css, pico.css, and water.css, and provides a snippet to easily add a theme to your website.
  55. Every programming languages paper since 1973, now searchable in one place — ▲ 38 · 💬 3 · 🔗 pldb.kirancodes.me
    A new site, pldb, indexes 33,580 programming languages papers from the first POPL in 1973 through this year's proceedings. It covers top-tier venues like POPL, PLDI, ICFP, and OOPSLA, plus dozens more, with abstracts for 26,107 papers. You can search by title, author, or keyword, filter by venue, year, or award winners, and even use advanced queries like author:"Xavier Leroy" or title:egraphs. Metadata comes from dblp, abstracts from SIGPLAN OpenTOC, OpenAlex, and Semantic Scholar.
  56. AI Model Opus 5.5 Uncovers a Lost 1615 Account of Dodo Hunting — ▲ 37 · 💬 2 · 🔗 resobscura.substack.com
    Historian Benjamin Breen describes using the AI model Opus 5.5 to search the Dutch East India Company archives, where it found a previously unnoticed 1615 ship's log that mentions catching dodos. The model also surfaced a possible new reference to the extinct red rail and a theory linking a Portuguese Jesuit's 1616 'ostrich' to the dodo painted for Mughal Emperor Jahangir. Breen argues AI excels at tireless archival searching but still needs human experts to ask the right questions and judge significance.
  57. Ink & Switch's PlayBook turns paper notebooks into programmable devices — ▲ 35 · 💬 7 · 🔗 youtube.com
    Ink & Switch has unveiled PlayBook, a programmable paper notebook that blends physical pages with digital interactivity. A Hacker News commenter calls it one of the coolest demos from the lab in a while, though notes the published live demo omits several demos shown in the talk and leaves unclear how users can build their own, making deeper evaluation difficult.
  58. Yantra - LALR(1) parser generator for C++ — ▲ 31 · 💬 16 · 🔗 github.com
    Yantra is a powerful LALR(1) parser generator for C++ that integrates a lexer, parser, and AST builder into one tool. It builds a full AST first, then walks it top-down, enabling parent actions before children. Features include UTF-8 support, multi-mode lexer, and optional amalgamated output. With no dependencies beyond the C++ standard library, it's easy to build and use. Ideal for compiler and language tooling, Yantra offers a fresh alternative to Bison and ANTLR.
  59. rhun - Assembly-powered code editor for speed — ▲ 28 · 💬 9 · 🔗 rhun.app
    rhun is a small, fast code editor written in assembly, designed to keep your coding environment light while AI handles the heavy lifting. It offers tabs, syntax highlighting, fuzzy search, and optional Vim mode, and runs on Windows, Linux, and macOS. With a focus on minimal weight and maximum flow, rhun eliminates bloat so you can stay in the zone. Install via terminal or download the release, and experience a responsive editor that prioritizes performance and simplicity.
  60. NASA Finds a 728-Foot Crater on the Moon, and It Could Reshape Plans for Future Bases — ▲ 27 · 💬 54 · 🔗 nytimes.com
    Astronomers spotted the largest fresh crater on the Moon since at least 2009: a 728-foot-wide basin near the Sea of Crises, carved in spring 2024 by an object 30 to 70 feet across. The impact released energy comparable to 15,500 tons of TNT, and such strikes happen only once every 132 years. The discovery, reported in two Science Advances studies, raises questions about the risks that impacts pose to planned human outposts.
  61. How a Hoax in England Delayed the Truth About Our African Origins — ▲ 18 · 💬 4 · 🔗 lflank.wordpress.com
    The story of Australopithecus africanus is intertwined with the Piltdown Man hoax, which convinced European scientists that humans evolved in Europe with large brains. When Raymond Dart discovered the Taung Child in South Africa in 1924, his claim that it was a human ancestor was dismissed. Only after Piltdown was exposed as a fraud in 1953 did the scientific community accept Africa as the cradle of humankind. Today, A. africanus is one of the best-known early hominins, with fossils from at least 250 individuals found in South Africa.
  62. Hardcore Coworking: 50 Minutes of Work, 10 Minutes of Break, No Excuses — ▲ 16 · 💬 5 · 🔗 cmart.blog
    The author proposes a strict coworking protocol to defeat procrastination through body doubling. Participants bring pre-planned tasks, work in 50-minute blocks with 10-minute breaks, and avoid all leisure activities. The group meets in public spaces like libraries or cafes, not homes, and does not linger after the final break. The protocol is open for feedback and revision after trial runs.
  63. NanoPing uses three Wi-Fi networks at once to stream live video with zero late packets — ▲ 15 · 💬 11 · 🔗 nanoping.com
    In a 30-meter lab walk, a laptop streamed 7 Mbps live video over three Wi-Fi networks simultaneously. While each network alone would have caused stutters, NanoPing delivered every packet within 100 ms, with no frames late or lost. The system dynamically steered traffic to the strongest network and used the others to repair losses, achieving 100% on-time delivery.
  64. Jevotron detects anomalies in your data with one command — ▲ 12 · 💬 1 · 🔗 cmungall.github.io
    Jevotron is a command-line tool that scans CSV, YAML, JSON, and other formats for anomalies using LLM guidance. It previews inputs, scores fields, and reuses cached assessments. On 24 public agent traces, it matched 79.8% of step-quality labels with 89.7% precision for harmful steps. Reports show probabilities and source locations, making it a practical review aid.
  65. Jacques Barzun: The Cultural Historian Who Saw Modern Art as a Deliberate Destroyer — ▲ 12 · 💬 5 · 🔗 csmonitor.com
    In a 1980 interview, Jacques Barzun argues that modern artists since 1920 have been purposeful destroyers, using shock and parody to break down artistic expectations built over 500 years. He sees this as a historical obligation at the end of a masterpiece-heavy era, and suggests electronic music might offer the only genuine new path forward.
  66. aweb gives AI agents persistent identities and wakes them when work arrives — ▲ 12 · 💬 5 · 🔗 aweb.ai
    aweb is an MIT-licensed, federated, self-hostable communication layer for AI agents. It provides stable identities, durable mail and chat, and wake-up events across sessions, runtimes, and machines. Agents can run in Claude Code or Pi, with maintained integrations that wake sessions when messages arrive. Independently operated servers can federate, and the CLI and API are runtime-independent. The system separates identity authority from message delivery via AWID, allowing organizations to retain control over team membership and signing keys.
  67. Claude rewrote my clipboard app for the Mac App Store — ▲ 9 · 💬 14 · 🔗 jefftk.com
    Jeff Kaufman's Clipboard Normalizer strips fonts, colors, and sizes from copied text, keeping only Markdown-compatible semantics. Now a free menu bar app on the Mac App Store, it can also convert rich text to Markdown and back. Claude handled the App Store rewrite: swapping Pandoc for SwiftSoup, writing Swift wrappers, merging two apps into one, adding tests, and walking him through review.
  68. Strutt's Powered Wheelchair Is Not a Wheelchair (For Legal Reasons) — ▲ 9 · 💬 15 · 🔗 core77.com
    Singaporean startup Strutt avoided the FDA's Class II medical device approval process by not labeling its ev¹ as a wheelchair. The four-motor vehicle features electronic stability control, anti-lock brakes, and self-driving modes, and can climb a 13° slope, clear a 2-inch curb, and cross a 4.7-inch gap. It costs $7,500, but insurance won't cover it and its 640 Wh battery isn't cleared for air travel.
  69. Effective altruism is this century's biggest idea — ▲ 9 · 💬 31 · 🔗 economist.com
    The Economist's leader argues that effective altruism, an ethical doctrine born at Oxford, has moved from academia to the corridors of power with remarkable speed. Its adherents populate Silicon Valley and lead top AI companies, while the Trump administration brands it 'defective altruism' and 'dystopian.' The movement is admirable, the article concludes, but dangerous.
  70. Incredible: A Rust TUI Framework for the 21st Century — ▲ 8 · 💬 12 · 🔗 incredible.rs
    Incredible is a new Rust framework for building text user interfaces. It's declarative, event-driven, and reactive, with a typed element tree and theming engine. It compiles to terminal, native window, or browser (WASM) from one codebase. Currently in private alpha, it supports modern terminal features and composable elements.
  71. Box3D's Inscribed Spheres Trick Slashes SAT Edge Testing — ▲ 8 · 💬 0 · 🔗 box2d.org
    The separating axis test (SAT) is robust for contact manifolds but can be O(N³) for polytopes. Box3D already uses SIMD, feature caching, and contact recycling, yet large piles still bottleneck on edge pair tests. A new technique uses inscribed spheres to compute a cheap upper bound on separation for faces and edges. The key innovation: using the Gauss map edge arc to cull edges before the expensive pair test, drastically reducing candidate pairs.
  72. Utility Locators Compete for $2,500 in the International Locate Rodeo — ▲ 7 · 💬 0 · 🔗 locaterodeo.net
    The 2026 International Utility Locate Rodeo takes place December 7–8 in Springfield, Missouri, and is free for competitors. Locators from four divisions—electric, gas, telecom, and water—compete for cash prizes, including $1,000 per division winner and $2,500 for the Locate From Hades champion. The event promotes damage prevention best practices and recognizes top locating performance.
添加评论
点赞收藏
点踩分享查看原文
评论
?
参与讨论