source&pool
A daily wire of long-form journalism, video, and discourse — filed, tagged, and laid out flat.
VOL. I·NO. 01
FRIDAY, SEPTEMBER 25, 2026
Hacker News3453X 主题热门3382CNBC64YahooFinance61aihot55Verge489to5Mac47IGN44MacRumors39Kotaku36TechCrunch27Engadget239to5Google22AndroidAuthority22NintendoLife20Eurogamer17ArsTechnica16Guardian16Wccftech15BusinessInsider14Investor'sBusinessDaily14USAToday14FoxBusiness13WarhammerCommunity13Polygon12PushSquare12TechPowerUp12Fortune11Gizmodo10SeekingAlpha10CBS9Mashable9NBC9NPR9VideoGamesChronicle9CNET8CNN8MotleyFool8Gematsu8GSMArena8NintendoEverything8Notebookcheck8VideoCardz8ABC7bgr7BleepingComputer7PureXbox7AppleInsider6CoinDesk6Fox6NewYorkPost6Yahoo6AlJazeera5DroidLife5HollywoodReporter5InsiderGaming5PlayStationLifeStyle5TechSpot5Tom'sGuide5WIRED5Aftermath4Deadline4GamesIndustry.biz4PetaPixel4SamMobile4SlashGear4Conversation4Hacker4UploadVR4WindowsCentral4404Media3AndroidCentral3AndroidPolice3BellofLostSouls3EventHubs3GameInformer3GearPatrol3Hackaday3HuffPost3Lifehacker3Motor13PCMag3PokeBeach3RockPaperShotgun3RPGSite3SouthChinaMorningPost3Variety3WhatHi-Fi?3WSB-TV36abcPhiladelphia2ABC7LosAngeles2AZFamily2Benzinga2ChromeUnboxed2Currently2DaringFireball2DCRainmaker2Futurism2GAMINGbible2HouseDigest2Jalopnik2MyNintendo2Nature2XBOXWire2Pokemon2qz2Road&Track2SeattleTimes2SFGATE2SimsCommunity2TimeExtension2TODAY2TweakTown224/7WallSt.180Level1ageofempires1Alternet1AndroidHeadlines1ArizonaSports1BostonGlobe1BusinessTimes1Yahoo!FinanceCanada1CalMatters1CarBuzz1cbn1ClaimDepot1ColoradoSun1Skin.ClubCommunity1consequence1ChristianScienceMonitor1DailyKos1DarkHorizons1Decrypt1Defense1denver71Designboom1DigitalCameraWorld1DigitalFoundry1DirtonDirt1Draftsim1DSOGaming1Electrek1empireonline1GameGPU1erictopol.substack1Fangoria1ForexFactory1franchisetimes1DetroitFreePress1GameRant1GameWorldObserver1GamingOnLinux1GeekWire1GeekyGadgets1Global1GosuGamers1Gothamist1Hackster.io1Hodinkee1HoustonChronicle1Independent1InsideEVs1InterestingEngineering1investor.costco1Invezz1iPhoneinCanada1LosAngelesTimes1MacObserver1MakeUseOf1Mashed1MLive1MorningBrew1MortgageDaily1Motorsport1MP1st1NBC5Chicago1BloombergLaw1SemiAnalysis1Newsshooter1Newsweek1NintendoWire1nrn1NYT1CrudeOilPricesToday1OneMileataTime1OregonPublicBroadcasting1OregonLive1PersonaCentral1Phoronix1politico.eu1QuantaMagazine1Realtor1RoadtoVR1RockstarINTEL1Salon1SeattleRed1Semafor1SanFranciscoChronicle1SimpleFlying1GhostHowls1Slate1SlippedDisc1SoraNews241SpaceNews1statnews1the5krunner1DailyBeast1DailyMeal1Drive1Hindu1Intercept1Times1TimesofIndia1TimesUnion1TMZ1TopGear1YahooFinanceUK1PCMagUK1Vulture1WCVB1WFMZ1WHYY1WKYT1YGOrganization1YourTango1
  1. 001Hacker NewsSEP · 23English

    GPT-6 Astra is the best vision model

    OpenAI's GPT-6 Astra vision model achieves state-of-the-art object detection performance, scoring 82.1% mAP@50 on Roboflow Vision Evals and outperforming competitors like Qwen3.8 Max and GPT-5.6 Sol. The model excels at computer vision tasks including object detection, visual reasoning, box prompting, and segmentation, combining fine-grained detail detection with semantic understanding for effective auto-annotation and classification across diverse visual scenarios.

    By Piotr Skalski
  2. 002Hacker NewsSEP · 22English

    Detecting and Tracking Flying Objects

    A startup needed reliable drone detection that could distinguish drones from birds and planes while tracking their location. The engineer built a real-time system that outperforms published state-of-the-art models on all object sizes by using a modern detection backbone, adding a high-resolution head for small distant objects, and fixing corrupted training data that caused false positives.

    By Denis Zenios
  3. 003Hacker NewsSEP · 19English

    What failed when I trained an object detector on synthetic images only

    A year-long project trained traffic-sign detectors using only synthetic rendered images, achieving 75–84% of the performance of models trained on real photographs. Fine-tuning the synthetic-pretrained model on real data improved detection by 0.15 mAP50, with the largest gains on underrepresented sign classes, though modeling snow proved ineffective due to lack of real snowy conditions in test data.

    By Stepan Galickov ·