Insizakalo yommeli wemakethe empunga eShayina iphinda ithengise ukufinyelela kumamodeli e-Claude AI ku-Anthropic ngentengo yeshumi, izuza ngeziqinisekiso ezibiwe, ukufaka esikhundleni samamodeli, kanye nokuthengiswa kabusha kwezicelo nezimpendulo zabasebenzisi, ngokusho kocwaningo olwenziwe ngumcwaningi we-Oxford China Policy Lab uZilan Qian.
Lawa manethiwekhi e-proxy, abathuthukisi baseShayina abawabiza ngokuthi “iziteshi zendlela,” asebenza obala nge-GitHub, i-Taobao, kanye ne-Telegram. Okutholakele kukaQian kuqinisekisa izixwayiso zakamuva ezivela e-White House kanye ne-Anthropic. Ekupheleni kuka-Ephreli, uhulumeni wase-US wasola izinhlangano zaseShayina ngokucwiliswa “kwezinga lezimboni”—ukuqeqesha amamodeli azo e-AI ngezimpendulo zikaClaude—ngamashumi ezinkulungwane zama-akhawunti e-proxy, kuyilapho i-Anthropic ithole cishe ama-akhawunti angu-24,000 axhumene nezinkampani ezintsha ze-AI i-DeepSeek, i-Moonshot AI, kanye ne-MiniMax emuva ngoFebhuwari.
UQian uchaze uhlelo lapho umhlanganyeli ngamunye enesibopho sesixhumanisi esisodwa noma ezimbili. Abahlinzeki bama-akhawunti ezingeni eliphezulu babhalisa ama-akhawunti ngobuningi mahhala ama-credits e-Anthropic angu-$5, baphinde bathengise ama-credits angasetshenziswanga kwamanye ama-akhawunti, basebenzise izaphulelo zezinkampani nezemfundo, noma bahlukanise okubhaliselwe kukaClaude Max okungu-$200 phakathi kwabasebenzisi abaningi ngemikhawulo yamathokheni ehora ngalinye. Amanye ama-akhawunti akhokhelwe ngamakhadi asebhange ebiwe aphelela echibini nge-inthanethi mahhala. Ukuze bagweme ukuqinisekiswa kobunikazi, abalamuli baya emazweni ase-Afrika naseLatin America futhi baqashe abantu ukuthi baqinisekiswe mathupha. Isibonelo esibhaliwe kwakuyimakethe emnyama ye-biometric i-Worldcoin, lapho ama-scan e-iris aqoqwe kubantu baseCambodia naseKenya athengiswa ngaphansi kwama-$30.
Ingozi ehlukile ihlotshaniswa nokufakwa esikhundleni kwamamodeli e-AI. Abacwaningi baseJalimane abavela e-CISPA Helmholtz Center for Information Security bahlole izinsizakalo ze-proxy eziyi-17 ezinjalo futhi bathola ukuthi imodeli ye-AI ekhangisiwe ivame ukungafani neyangempela. Ukufinyelela, okumakethwa njenge-"Gemini-2.5," kuthole amaphuzu angu-37% ekuhlolweni kwezokwelapha, kuyilapho i-API esemthethweni ithole cishe u-84%. Ngokusho kukaQian, esikhundleni se-Claude Opus eceliwe, abasebenzisi bangathola izimpendulo ezivela ezinjinini ze-Sonnet, Haiku, noma ze-AI zaseShayina ezishibhile.
Abaqhubi be-proxy nabo baqopha zonke izicelo nezimpendulo ezidlula kumaseva abo. Abathuthukisi abaningana baseShayina batshele uQian ukuthi i-premium yokufinyelela empeleni iyindlela yokuheha amakhasimende, kuyilapho imodeli yebhizinisi yangempela yakhelwe ekuqoqweni kwedatha. I-HuggingFace isivele isingatha amasethi edatha e-Claude Opus 4.6 yokucabanga okungaziwa ukuthi ivelaphi. Idatha enjalo ibaluleke kakhulu ekuhluzweni, njengoba izimpendulo zokucabanga zingaqoqwa ngokuhlelekile futhi zisetshenziswe ukuqeqesha amamodeli ancintisanayo. Amaseva e-proxy ahlinzeka ngomfudlana wedatha ofanayo ngomzamo omncane: abasebenzisi abakhokhayo bakha izinto zokuqeqesha ngokwabo.
Kodwa-ke, usongo alugcini nje ekuqeqeshweni kwamamodeli. Abathuthukisi bavame ukwabelana ngezingcezu zekhodi yobunikazi, isakhiwo se-API, kanye ne-logic yokuqinisekisa nama-ejenti e-AI. Uma le thrafikhi idlula ku-proxy engathembekile, inkampani ithumela idatha yangaphakathi kuseva yangaphandle ngaphandle kwesibopho sokuyicubungula. Ingozi efanayo yavela ngo-2023 kwa-Samsung, lapho onjiniyela bayo bethumela ikhodi yomthombo ku-ChatGPT, ngaleyo ndlela bedalula idatha ebucayi yokukhiqiza i-semiconductor kumaseva e-OpenAI.
I-Anthropic yavimba ukufinyelela kuClaude kwezinhlangano ezilawulwa yiShayina ngoSepthemba futhi kamuva yaqinisa ukuqinisekiswa komsebenzisi. Kodwa-ke, ucwaningo lukaQian lubonisa ukuthi isinyathelo ngasinye esisha sakha imakethe ehlukile yokugwema kunokuvimba ukufinyelela okungagunyaziwe.
Source:
Source: 3dnews.ru
