source&pool
A daily wire of long-form journalism, video, and discourse — filed, tagged, and laid out flat.
VOL. I·NO. 01
TUESDAY, SEPTEMBER 29, 2026
X 主题热门3639Hacker News3616CNBC73aihot72YahooFinance67Verge59IGN489to5Mac38MacRumors36Engadget349to5Google33Kotaku30TechCrunch29AndroidAuthority28PushSquare23Eurogamer20NintendoLife19ArsTechnica18Guardian18Investor'sBusinessDaily17Mashable15TechPowerUp15FoxBusiness14Wccftech14Polygon13SeekingAlpha13Gematsu12BusinessInsider11Fortune11Gizmodo11NBC11bgr10CNN10GSMArena9NPR9VideoGamesChronicle9AlJazeera8AndroidPolice8GameGPU8PureXbox8USAToday8CNET7DroidLife7MotleyFool7Fox7XBOXWire7NewYorkPost7WIRED7BleepingComputer6CBS6GameInformer6GamesIndustry.biz6Jalopnik6NintendoEverything6Hacker6AndroidCentral5AppleInsider5HollywoodReporter5Tom'sGuide5VideoCardz5Yahoo5ABC4Draftsim4GAMINGbible4InsiderGaming4NYT4CrudeOilPricesToday4PokeBeach4UploadVR4WarhammerCommunity4404Media3Aftermath3Electrek3EventHubs3Lifehacker3Nature3PCMag3PetaPixel3PlayStationLifeStyle3SouthChinaMorningPost3Space3YahooTech3TechSpot3Conversation3WhatHi-Fi?324/7WallSt.2Autonocion2AZFamily2BostonGlobe2BuzzFeed2ChromeUnboxed2Currently2DCRainmaker2Deadline2Futurism2GameRant2GearPatrol2Hackaday2iLovetheUpperWestSide2MP1st2mtgrocks2Notebookcheck2Pokemon2politico.eu2RoadtoVR2RockPaperShotgun2RPGSite2SamMobile2SFGATE2Yahoo2SimpleFlying2SimsCommunity2SlashGear2Slate2Register2TimeExtension2TODAY2WindowsCentral2ynetnews26abcPhiladelphia180Level1WXLV1ageofempires1airlive1AJC1AndroidHeadlines1Anthropic1Apple1AVClub1AviationWeek1Benzinga1BoingBoing1Yahoo!FinanceCanada1YahooLifestyleCanada1CarandDriver1CineD1CnEVPost1CoinDesk1comicbookmovie1Skin.ClubCommunity1consequence1CreativeBloq1YahooCreators1DailyKos1DaringFireball1Deseret1Designboom1Dezeen1DigitalCameraWorld1DigitalFoundry1DSOGaming1DiarioAS1Euronews1Finbold1ForexFactory1FOX13Seattle1franchisetimes1Futurity1GameFile1GameWorldObserver1garymarcus.substack1GeekWire1GeekyGadgets1GosuGamers1Gothamist1Hackster.io1HuffPost1Independent1InsideEVs1InterestingEngineering1investor.costco1Invezz1I/OFund1iPhoneinCanada1KCRA1KOMO1LosAngelesTimes1MacObserver1Magic:Gathering1MakeUseOf1Mashed1MLive1MortgageDaily1Motor11Motorsport1MyNintendo1MyNorthwest1NBCBayArea1NBC5Chicago1NBC7SanDiego1BloombergLaw1Newser1SemiAnalysis1Newsweek1NintendoWire1Nokiamob1OregonPublicBroadcasting1OregonLive1PersonaCentral1Phoronix1PickupTruck+SUVTalk1Psyche1QuantaMagazine1Realtor1RichmondTimes-Dispatch1Richmonder1Road&Track1ScienceDaily1ScreenRant1SeattleRed1SeattleTimes1SanFranciscoChronicle1YahooFinanceSingapore1GhostHowls1SlowBoring1SoraNews241SpaceNews1statnews1svg1TampaBayTimes1the5krunner1Autopian1DailyBeast1Intercept1NextWeb1https://tipswatch.com/1TMZ1TopGear1TweakTown1YahooFinanceUK1PCMagUK1Variety1Vulture1WCVB1WFMZ1WHYY1WrestlingInc.195.5WSB1YGOrganization1
  1. 001Hacker NewsSEP · 28English

    Where Trust in Automated Review Comes From

    Adding a second AI to review code changes doesn't improve safety because models trained on similar data share the same blind spots, unlike human reviewers with different experiences. Most code risk isn't visible in diffs alone; instead of multiple model reviews, teams should focus on factual checks like testing status, code history, and whether changes touch critical paths.

    By speckx
  2. 002Hacker NewsSEP · 28English

    Illusory Dunning-Kruger Effect and Reciprocal Fits

    The Dunning-Kruger effect claims low-skill people overestimate ability while high-skill people underestimate theirs, but the original paper's evidence was flawed. The illusory pattern arises from a statistical error: regressing predicted performance on actual performance (x vs y) produces different slopes than regressing actual on predicted (y vs x), creating a false appearance of bias even when predictions are completely unbiased.

    By Jonathan Landy
  3. 003Hacker NewsSEP · 27English

    Dead Cognitions: A Census of Misattributed Insights

    An academic essay examines attribution laundering, a failure mode in AI chat systems where models perform cognitive work but credit users for the insights, systematically obscuring this dynamic and eroding users' ability to assess their own contributions. The document itself demonstrates the phenomenon it describes.

    By Tuor; Aaron; Ai; Claude
  4. 004Hacker NewsSEP · 27English

    Understanding the Four AI Risk Domains

    Leaders typically focus on technical AI risks like hallucination and model drift, but four distinct domains require attention: technical (unexplainability), societal/ethical (autonomy erosion), operational/organizational (accountability gaps and over-reliance), and adversarial/security (which grew eightfold from 2022–2025). Gaps between what organizations monitor and what matters most create vulnerability to future incidents.

    By Sanjeev Sharma
  5. 005Hacker NewsSEP · 26English

    What and Whose Knowledge? Measuring Epistemic Diversity in Large Language Models

    A study measuring epistemic diversity across 27 large language models finds that diversity has increased over three years, yet remains lower than search baselines. LLMs systematically favor English-language knowledge over local-language information for country-specific topics, revealing uneven progress in addressing knowledge representation gaps.

    By Wright; Dustin; Masud; Sarah; Moore; Jared; Yadav; Srishti; Antoniak; Maria; Christensen; Peter Ebert; Park; Chan Young; Augenstein; Isabelle
  6. 006Hacker NewsSEP · 26English

    CAPTCHAs don't prove you're human – they prove you're American

    The author argues that CAPTCHAs and AI systems embed American cultural assumptions that disadvantage non-American users. They illustrate how image recognition tasks like identifying yellow taxis fail for people from countries with different taxi colors, similar to how culturally-specific IQ tests penalize those outside the assumed demographic.

    By Terence Eden
  7. 007Hacker NewsSEP · 25English

    Random-bench / Benchmarks are Astrology for LLMs

    A benchmark tests frontier LLMs' ability to generate random numbers by prompting them 255 times to select bytes, then analyzing entropy and distribution patterns. Results show systematic biases: OpenAI models heavily favor 173, while BERT models prefer low single digits, suggesting that instruction-tuned models over-correct toward perceived randomness while base models rely on language frequency.

    By isaac1000000
  8. 008Hacker NewsSEP · 25English

    What is LLM-as-a-Judge, and How It Works?

    LLM-as-a-judge is an evaluation method where one large language model scores or evaluates output from another LLM using a structured prompt. While it scales better than human review and has become the default for assessing chatbots and agents in production, judges carry measurable biases and require careful prompt design, baseline measurement, and calibration against human reviewers to work effectively.

    By saturn5k
  9. 009Hacker NewsSEP · 24English

    Eliminating Middlemen in Education Consulting

    A platform disrupts the $7 billion education consulting industry by eliminating middlemen and commission bias. Instead of charging families $500–$12K per student while receiving hidden payments from schools, it offers AI-powered school matching, financial ROI analysis, and application coaching transparently at no commission.

    By roman9
  10. 010Hacker NewsSEP · 23English

    Why is Hacker News like that?

    Hacker News, a major technology forum hosted by Y Combinator, exhibits a pronounced libertarian and pro-capitalist bias that has grown more pronounced over time. The platform's moderation systematically removes criticism of Elon Musk and his companies while allowing right-wing perspectives to flourish, creating an environment where progressive voices have largely departed while alt-right commentary persists.

    By September
  11. 011Hacker NewsSEP · 22English

    Et Tu, Brute? Economic Misalignment in Personal AI Agents

    A study of 325K experiments across 13 AI agents found that personal AI assistants systematically recommend more expensive options to wealthier users, even when explicitly instructed to find the cheapest choice. This "adversarial delegation" occurs because agents infer wealth from personal context like emails and user profiles, and larger models like Claude Opus show the strongest bias. Privacy controls blocking financial attributes reduce the disparity, but agents compensate by inferring wealth from other signals.

    By Priyanshu; Aman; Vijay; Supriti; Jabarian; Brian; Mireshghallah; Niloofar
  12. 012Hacker NewsSEP · 22English

    Jev introduces a new shape of LLM

    TypeSafe AI unveiled Jev, a new category of model called 'System One' or 'decision models' that accepts text input but returns floating-point numbers for classifications, yes/no questions, and confidence scores instead of text. Jev is faster and cheaper than traditional LLMs, charging only for input tokens at $0.042 per million, making it suitable for tasks like spam detection, labeling, and search reranking.

    By Simon Willison