Inkonzo yommeli wemarike engwevu eTshayina iphinda ithengise ukufikelela kwiimodeli zeClaude AI kwi-Anthropic ngexabiso eliphantsi, ifumana inzuzo kwiinkcukacha ezibiweyo, ukutshintsha iimodeli, kunye nokuthengiswa kwakhona kwezicelo zabasebenzisi kunye neempendulo, ngokutsho kophando olwenziwe ngumphandi we-Oxford China Policy Lab uZilan Qian.
Ezi nethiwekhi zeproxy, abaphuhlisi baseTshayina abazibiza ngokuba "zizikhululo zendlela," zisebenza ngokukhululekileyo ngeGitHub, iTaobao, kunye neTelegram. Iziphumo zikaQian ziqinisekisa izilumkiso zamva nje ezivela kwiWhite House kunye neAnthropic. Ekupheleni kuka-Epreli, ulawulo lwase-US lutyhola amaziko aseTshayina ngokuhluza "izixhobo zoshishino" - eqeqesha iimodeli zawo ze-AI kwiimpendulo zikaClaude - ngokusebenzisa amashumi amawaka eeakhawunti zeproxy, ngelixa i-Anthropic ichonge malunga neeakhawunti ezingama-24,000 ezinxulumene neenkampani ezintsha ze-AI iDeepSeek, iMoonshot AI, kunye neMiniMax ngoFebruwari.
UQian uchaze icebo apho umthathi-nxaxheba ngamnye enoxanduva lwekhonkco elinye okanye amabini. Ababoneleli beeakhawunti kwinqanaba eliphezulu babhalisa iiakhawunti ngobuninzi ukuze bafumane iikhredithi ze-Anthropic ze-$5 simahla, baphinde bathengise iikhredithi ezingasetyenziswanga kwezinye iiakhawunti, basebenzise izaphulelo zenkampani nezemfundo, okanye bahlule ubhaliso lwe-Claude Max lwe-$200 phakathi kwabasebenzisi abaninzi ngokusebenzisa imida ye-token yeyure. Ezinye iiakhawunti ezihlawulelwe ngamakhadi ebhanki abiweyo ziphelela echibini simahla. Ukuze baphephe ukuqinisekiswa kobuwena, abalamli baya kumazwe aseAfrika naseLatin America kwaye baqeshe abantu ukuba baqinisekiswe ngokwabo. Umzekelo obhaliweyo yayiyi-biometric black market Worldcoin, apho ii-iris scans eziqokelelwe kubemi baseCambodia naseKenya zithengiswa ngaphantsi kwe-$30.
Umngcipheko owahlukileyo unxulunyaniswa nokutshintshwa kweemodeli ze-AI. Abaphandi baseJamani abavela kwiZiko le-CISPA Helmholtz loKhuseleko loLwazi bavavanye iinkonzo ze-proxy ezili-17 baza bafumanisa ukuba imodeli ye-AI ebhengeziweyo idla ngokungahambelani neyokwenyani. Ukufikelela, okuthengiswe njenge-"Gemini-2.5," kufumene amanqaku angama-37% kuvavanyo lwezonyango, ngelixa i-API esemthethweni ifumene phantse ama-84%. Ngokutsho kukaQian, endaweni yeClaude Opus eceliweyo, abasebenzisi banokufumana iimpendulo ezivela kwiinjini ze-Sonnet, iHaiku, okanye i-AI zaseTshayina ezingabizi kakhulu.
Abaqhubi beeproksi barekhoda zonke izicelo kunye neempendulo ezidlula kwiiseva zabo. Abaphuhlisi abaninzi baseTshayina baxelele uQian ukuba iprimiyamu yokufikelela ngokuyintloko yindlela yokutsala abathengi, ngelixa imodeli yokwenyani yeshishini yakhelwe ekuqokeleleni idatha. I-HuggingFace sele igcina iiseti zedatha zeClaude Opus 4.6 yokuqiqa engaziwayo imvelaphi. Olu lwazi luxabiseke kakhulu kwi-distillation, njengoko iimpendulo zokuqiqa zinokuqokelelwa ngokucwangcisiweyo kwaye zisetyenziswe ukuqeqesha iimodeli ezikhuphisanayo. Iiseva zeeproksi zibonelela ngomjelo wedatha ofanayo ngomzamo omncinci: abasebenzisi abahlawulayo benza izixhobo zoqeqesho ngokwabo.
Nangona kunjalo, isoyikiso asipheleli nje kuqeqesho lweemodeli. Abaphuhlisi badla ngokwabelana ngeziqwenga zekhowudi yobunini, isakhiwo se-API, kunye ne-logic yokuqinisekisa kunye neearhente ze-AI. Ukuba olu hambo ludlula kwi-proxy engathembekanga, inkampani ithumela idatha yangaphakathi kwiseva yomntu wesithathu ngaphandle kwesibophelelo sokuyicubungula. Umngcipheko ofanayo wavela ngo-2023 kwi-Samsung, xa iinjineli zayo zangenisa ikhowudi yomthombo kwi-ChatGPT, ngaloo ndlela zityhila idatha yokwenziwa kwe-semiconductor ebucayi kwiiseva ze-OpenAI.
I-Anthropic yavala ukufikelela kwiClaude kwiinkampani ezilawulwa yiTshayina ngoSeptemba yaza emva koko yaqinisa ukuqinisekiswa kwabasebenzisi. Nangona kunjalo, uphando lukaQian lubonisa ukuba indlela nganye entsha yadala imakethi eyahlukileyo yokuthintela endaweni yokuthintela ukufikelela okungagunyaziswanga.
umthombo:
umthombo: 3dnews.ru
