source&pool
A daily wire of long-form journalism, video, and discourse — filed, tagged, and laid out flat.
VOL. I·NO. 01
WEDNESDAY, OCTOBER 7, 2026
Hacker News1412X 主题热门1376CNBC104PANews77Channel News Asia56ChainCatcher51Coinness42SouthChinaMorningPost35YahooFinance32Crypto.news30Verge21Cointelegraph20aihot189to5Google17Decrypt17Variety16CoinDesk15IGN15MacRumors15Guardian14Engadget13TechCrunch13FoxBusiness129to5Mac11Kotaku11ArsTechnica10Eurogamer9NBC9DW8Gematsu8Gizmodo8BitcoinMagazine7NintendoLife6Polygon6AMBCrypto5BBC World5Deadline5Investor'sBusinessDaily5PushSquare5TechPowerUp5DigitalFoundry4GSMArena4Motor14XBOXWire4NPR4SeekingAlpha4WarhammerCommunity4Wccftech4AndroidAuthority3Fox3GameInformer3NintendoEverything3Notebookcheck3USAToday3VideoCardz3ABC2AlJazeera2AndroidPolice2AppleInsider2BusinessInsider2CBS2CNET2CNN2EIA2Electrek2FierceBiotech2GamesIndustry.biz2GAMINGbible2LosAngelesTimes2Nature2Pokemon2politico.eu2SEC2TechSpot2Drive2Hacker2Register2Tom'sGuide2VideoGamesChronicle2WindowsCentral2WIRED2404Media180Level1AboutAmazon1Aftermath1AFTVnews1AUTOMATON1AWealthofCommonSense1Barron's1bgr1BikeRadar1BleacherReport1BleepingComputer1BlockClubChicago1BuzzFeed1Carscoops1SanDiegoCountyCenter1Draftsim1DroidLife1Endpoints1Euronews1Federal Reserve1Fortune1Futurism1GamesBeat1GearPatrol1Hackaday1Hodinkee1HollywoodReporter1How-ToGeek1Independent1KOMO1Livingetc1magic.gg1Mashable1MyNorthwest1NoemaMagazine1NewYorkPost1NYT1CrudeOilPricesToday1PCWorld1PetaPixel1Phoronix1PokeBeach1PokémonGOHub1PureXbox1qz1RoadtoVR1RockPaperShotgun1SamMobile1ScreenRant1SeattleTimes1SearchEngineRoundtable1Yahoo1SlashGear1Slate1statnews1MobileReport1TweakTown1WeatherChannel1WhatHi-Fi?1WWD1ZDNET1
  1. 041Hacker NewsOCT · 02English

    Don't be fooled–LLMs don't reason

    An article argues that large language models lack true reasoning despite appearing intelligent. Using AlphaGo's famous Move 37 as an example of genuine reasoning—combining intuition with deliberate search through possible futures—the author contends that today's LLMs only perform pattern completion through token prediction, not authentic reasoning like humans or AlphaGo employ.

    By Thore Graepel
  2. 042Hacker NewsOCT · 02English

    Boston Dynamics Robot Hands

    Boston Dynamics developed a new 13-degree-of-freedom hand for its Atlas humanoid robot, designed to manipulate objects with dexterity comparable to human hands while balancing strength, ruggedness, and manufacturability for industrial work. The four-fingered design with an opposable thumb uses direct actuation and tactile sensing to enable complex manipulation tasks including tool use, prioritizing function over replicating human anatomy.

    By Mrodin
  3. 043Hacker NewsOCT · 02English

    Don't Look Up

    The author reflects on how computers evolved from programmer-friendly devices to consumer-focused tools, and proposes a future where AI serves as the primary interface while users retain full control and portability of their data through a decentralized base layer, allowing seamless transitions between AI environments.

    By Auke van Slooten
  4. 044Hacker NewsOCT · 02English

    Anyone Can Build an App Now, but Who Is Going to Pay for It?

    AI tools like Claude Code enable rapid app development, but building a product is only valuable once customers pay for it repeatedly. Startups must focus on customer acquisition and retention rather than perfecting the initial product, since competitors can replicate the app quickly but not the established customer base.

    By Vincent Schmalbach
  5. 045Hacker NewsOCT · 02English

    What MkDocs 2.0 means for your documentation projects

    MkDocs 2.0, a ground-up rewrite announced three weeks ago, introduces breaking changes including incompatibility with Material for MkDocs, removal of the plugin system, a rewritten theming system that limits advanced navigation features, a closed contribution model, and a shift from YAML to TOML configuration. The maintainers of Material for MkDocs present these findings and introduce Zensical, a new static site generator compatible with MkDocs 1.x, as an alternative.

    By Martin Donath
  6. 046Hacker NewsOCT · 02English

    Show HN: Hotel Lobby AI Video Generator – Two Photos to Rap Duet

    Hotel Lobby AI Video Generator lets users create short two-person rap performance videos by uploading two photos, which are swapped into the original choreography and music. HD videos cost 40 credits per second with generation times of 5–20 minutes depending on length, and the tool requires Google sign-in but is independent of the original COLORS brand.

    By jerry5678
  7. 047Hacker NewsOCT · 02English

    Slowly, then all at once: AI agents that know how a company works

    IonQ engineer built MindRoom, an open-source AI agent platform that works like a chat app, giving employees personal and team agents connected to company data sources like Slack, Jira, and code repositories. After initial slow adoption through pilots, a company-wide demo showing the agent's ability to explain complex programs drove rapid adoption, with hundreds signing up and adoption reaching about a third of eligible employees within weeks.

    By Bas Nijholt
  8. 048Hacker NewsOCT · 02English

    The Looming Closure of China's AI Frontier

    China's AI strategy is shifting toward restricting access to powerful AI models amid security concerns, even as Beijing promotes open-source development. Recent model releases capable of aiding cyberattacks and terrorism have prompted security officials to call for stricter oversight, while Chinese AI labs increasingly delay public releases and implement restrictive licensing terms.

    By Ryan Fedasiuk; Alexander Rosenquist
  9. 049Hacker NewsOCT · 02English

    Show HN: Wrapper – open a live terminal session from your phone

    Wrapper is a tool that lets you share a live terminal session from your computer to viewers on their phones via a secure WebRTC connection, with granular control over who can type. The terminal stays on your host machine, sharing is initiated with a keyboard shortcut that generates an expiring code, and all data is encrypted end-to-end with no metadata stored on relays.

    By icanvardar
  10. 050Hacker NewsOCT · 02English

    Nvidia/OpenShell: Safe, private runtime for autonomous AI agents

    OpenShell is Nvidia's safe, private runtime for autonomous AI agents that enforces security policies through kernel-level isolation and formal verification. Agents can access files, packages, APIs, and credentials within declared policy boundaries, with kernel controls and credential injection preventing unrestricted data access. The 0.1.x release adds stable cadence, isolation primitives, and expanded extensibility for Linux, macOS, and Windows with WSL 2.

    By NVIDIA
  11. 051Hacker NewsOCT · 02English

    Researchers Drove a Chinese EV into an Underground Mine. It Kept Calling China

    Norwegian researchers monitored a Chinese NIO ES8 electric SUV in an underground mine and found approximately 90% of its outbound connections reached servers in China, contradicting NIO's claims about local data processing in Europe. A parallel study by Norway's Ruter transport operator discovered that Chinese-made Yutong buses had remote access to battery and power systems, presenting potential security risks, while Dutch-made buses lacked such connectivity.

    By Rex Edison
  12. 052Hacker NewsOCT · 02English

    Nobody Is Thinking Anymore

    An opinion piece criticizing over-reliance on AI in professional settings, arguing that outsourcing thinking to language models undermines human judgment, decision-making, and skill development. The author advocates for preserving cognitive work and viewing AI as a tool to support rather than replace human thinking.

    By Yusuf Aytas
  13. 053Hacker NewsOCT · 02English

    GamesByAI – 528 browser games made with AI, and which tools made them

    A catalog analysis of 528 AI-generated browser games reveals Claude leads with 238 mentions, Claude Code is used in 38% of games, and Three.js powers 58%. Most games were created for game jams, with Action being the dominant genre at 41%, and creators report AI writing varying amounts of code.

    By kvickan
  14. 054Hacker NewsOCT · 02English

    Benchmarking retrieval for agents on messy real-world company knowledge

    Kapa introduced Company Knowledge Bench, a dataset of 1,000 real-world evaluation cases to measure retrieval performance for agents accessing company knowledge like documentation and chat messages. The benchmark reveals that optimized agentic retrieval (Kapa Deep) achieves 0.65 accuracy in five seconds, outperforming frontier models using basic grep matching at 0.61 but taking five times longer.

    By emil_sorensen
  15. 055Hacker NewsOCT · 02English

    Claude-Shaped Science

    Prof. Matthew Schwartz describes BootLoops, a toolkit designed to align AI capabilities with scientific research needs. By identifying "Claude-shaped" problems—quantitative challenges suited to LLM strengths—Schwartz built an open-source harness that helped Claude discover connections across physics, ecology, genetics, and other fields, addressing the impedance mismatch between what scientists need and what current AI does well.

    By famouswaffles
  16. 056Hacker NewsOCT · 02English

    Show HN: Noite – Enjoy Most of Cloudflare Worker APIs on Your Potato VPS

    Noite is an open-source, self-hostable platform that brings Cloudflare Workers APIs to personal VPS servers. It enables developers to deploy apps with durable objects, databases, and storage using a single command and git push workflow, maintaining full data ownership.

    By ryuzyy
  17. 057Hacker NewsOCT · 02English

    Wikipedia: List of Citogenesis Incidents

    Citogenesis is the creation of false information through circular reporting where inaccurate Wikipedia entries are copied by media outlets, then those media sources are cited back in Wikipedia as proof. The article documents well-known cases including fake deities, fabricated names, false biographical claims, and invented historical facts that persisted for years after being amplified through mainstream sources.

    By thunderbong
  18. 058Hacker NewsOCT · 02English

    Why Isn't AI Adoption Showing Up in Your P&L?

    Most organizations measure AI adoption by usage counts rather than actual value capture. A 2026 study of 22,000 developers reveals that AI-accelerated development shifts bottlenecks downstream—from code review to validation to operations—without increasing overall organizational velocity. True adoption metrics should track whether AI-built artifacts see real use beyond their creators, and organizations must implement cost discipline to manage token consumption.

    By Leah Brown
  19. 059Hacker NewsOCT · 02English

    SQLite-Vector is now free and open-source

    SQLite-Vector is now open-source and free, offering production-grade vector search capabilities as a SQLite extension. It supports multiple data types, SIMD acceleration, and TurboQuant quantization, running efficiently across mobile, browser, edge, and server platforms with minimal memory overhead.

    By Sqliteai
  20. 060Hacker NewsOCT · 02English

    Trading off compute for memory with activation checkpointing

    Activation checkpointing trades memory for compute by discarding intermediate activations during forward passes and recomputing them during backpropagation. Arsh Koneru, an ML infrastructure intern, was tasked with building an improved activation checkpointing scheme that minimizes extra compute while staying within a given memory budget, building on PyTorch's existing tools like torch.compile's AOT Autograd and min-cut optimization.

    By James Somers