🧠 Inhloko ye-Google DeepMind ithi ukuthuthukiswa kwe-AI kungafinyelela eqophelweni elibi maduze, nansi isizathu ↗
Inhloko ye-DeepMind iphakamisa ifulegi elincane eliphuzi: intuthuko ingase ingahlali ibushelelezi, iqondile ukusuka lapha. Icala liwukuthi "ukunqoba okulula" okuvela ekukhuleni kungaqala ukuncipha, futhi izithiyo zingase zishintshe kusukela kuma-GPU ziye kudatha, amandla, ukuhlolwa ... amapayipi angabonakali.
Akuyona into engenakwenzeka, impela - kuyisikhumbuzo sokuthi singase sidinge amaqhinga amasha, hhayi nje izando ezinkulu. Lokho kuzwakala kuqabula futhi kujabulisa kancane, kuye ngamazinga akho e-caffeine.
🔊 Amadivayisi e-OpenAI athuthukisa i-AI afaka phakathi izipikha ezihlakaniphile, imibiko yolwazi ↗
Kubikwa ukuthi i-OpenAI ibheke kwihadiwe, iqala ngedivayisi ehlakaniphile yesitayela sesipikha engahlala ebangeni lentengo yabathengi eliphakathi nendawo. Ushintsho - kuthiwa ifaka ikhamera, ngakho-ke akuyona nje "umsizi wezwi, kodwa futhi inomsindo omkhulu"; isondele "kumuntu ohlala naye osebenzisa i-AI" ... okuzwakala kuwusizo futhi kukhathaza kancane emoyeni ofanayo.
Kukhona futhi izingxoxo mayelana nezinye izigaba zamadivayisi eziqhubekayo, kodwa umqondo wesikhathi esifushane uwukuthi: I-OpenAI ifuna umzila oqondile empilweni yakho yansuku zonke ongahlungiwe ngeplatifomu yomunye umuntu. Ukuthathwa komhlaba okuncane, iphupho lomkhiqizo omncane - kokubili kungaba yiqiniso.
💰 Kubikwa ukuthi iNvidia ihlela ukutshala imali engu-$30 billion emzuliswaneni olandelayo wokuxhasa ngezimali we-OpenAI ↗
I-Nvidia ne-OpenAI bangase bahlele esinye isivumelwano esikhulu, kanti kubikwa ukuthi i-Nvidia ibheke inombolo yokutshalwa kwezimali engenangqondo ngempela emzuliswaneni olandelayo we-OpenAI. Inkulumo yokulinganisa ecashile imnandi... njengobuchopho bakho buzama ukubona inombolo bese iphenduka inkungu.
Okuthakazelisayo ukuthi i-strategic circular iyindlela ejikelezayo: inkampani ethengisa amafosholo ifuna isabelo esikhulu emayini yegolide. Noma mhlawumbe sekuphambene nalokho manje - imayini yegolide ithenga ilungelo lamafosholo engeziwe, kanti umenzi wefosholo naye ufuna ingxenye yalokho. Inhle kakhulu, ngendlela ye-capitalism yamuva.
🛡️ I-Anthropic Iqala Ukuphepha Kwekhodi KaClaude Kokuskena Ikhodi Esebenzisa I-AI ↗
I-Anthropic ishintshele endleleni ewusizo kakhulu: ukuskena kokuphepha okusizwa yi-AI okubheka ubuthakathaka kuma-codebase wangempela, bese kusikisela ama-patches abantu okufanele bawabukeze. Kubonakala sengathi kugcizelelwa "njengomuntu ngaphakathi kolwazi," ngoba akekho ofuna i-bot eqiniseka ngokweqile "elungisa" umkhiqizo buthule futhi iguqule uhlelo lwakho lokusebenza lube ubuciko obungabonakali.
Ingqikithi enkulu ukusheshisa ukuzivikela: uma abahlaseli bengasebenzisa i-AI ukuthola amaphutha ngokushesha, abavikeli badinga nenkinobho yabo ye-turbo. Kufana nokunikeza abantu abalungile izibuko zokubona ebusuku... futhi benethemba lokuthi amabhethri ngeke anyamalale ngendlela engaqondakali.
🎮 I-CEO entsha yemidlalo ye-Microsoft ifunga ukuthi ngeke igcwalise uhlelo lwe-ecosystem 'nge-AI slop engapheli' ↗
Ushintsho lobuholi bemidlalo ye-Microsoft luza nesithembiso sesimanje kakhulu: yebo, i-AI izovela, kodwa cha, abazami ukuminza abadlali odakeni lokuqukethwe kwe-algorithmic. I-“Endless AI slop” iyinkulumo enembe kakhulu futhi ebuhlungu cishe ebuhlungu - njengokuchaza ama-noodle asheshayo “njengezinti zosawoti ezingenamkhawulo.”
Isignali lapha iwukubeka: I-Microsoft ifuna ukubonakala njengeyokuqala kokunambitha, hhayi nje ukukala kuqala. Ukuthi lokho kuyasinda yini uma izimboni zokuqukethwe ziqala ukucula kusazobonakala.
🕹️ Isikhulu se-Xbox uPhil Spencer uyashiya iMicrosoft ↗
Ukuphuma kukaPhil Spencer kuyisikhathi esibalulekile sesiko le-Xbox, futhi ukusethwa komlandeli kuhambisana kakhulu ne-AI, kanye nobuholi obuvela ohlangothini lwe-CoreAI lwe-Microsoft. Lokho akusho ngokuzenzakalelayo ukuthi “imidlalo ye-AI yonke indawo,” kodwa kushukumisela inaliti ku-“AI izoba yingxenye yengxoxo yesu, njalo.”
Isikhathi sikaSpencer sasimayelana nokwakha kabusha okuqhubekayo kanye nokubheja okukhulu - manje kubonakala sengathi isikhathi esilandelayo singase sibe mayelana nokulungisa lokho ngokuzenzakalelayo, ukwenza kube ngokwakho, nanoma yini "ikusasa lokudlala" eliba yiyo lapho amamodeli ethola amathuluzi.
Imibuzo Evame Ukubuzwa
Kusho ukuthini uma inhloko ye-DeepMind ithi intuthuko ye-AI ingafika eqophelweni elibi?
“Iphuzu lokuminyana kwentuthuko ye-AI” libhekisela ekutheni izinzuzo eziqondile zanamuhla ezivela kumamodeli okukala nje zingase ziqale ukuncipha. Esikhundleni sokuthi ama-GPU abe yiwona kuphela umkhawulo, imikhawulo ingashintshela ekhwalithini yedatha, izindleko zamandla, noma ubunzima bokuhlola inqubekela phambili ngokuzethemba. Akuyona ngempela ukwehla okuhlala njalo - kuyisibonakaliso sokuthi ukukhuphuka okulandelayo kungadinga izindlela ezahlukene, hhayi nje amaqoqo amakhulu.
Yiziphi izinkinga ezingase zithathe indawo yama-GPU uma ukuthuthukiswa kwe-AI kufinyelela eqophelweni lokuminyana?
Uma kuvela iphuzu lokuminyana kwentuthuko ye-AI, "ukufaka amapayipi" kuba yisihloko: ukuvikela idatha esezingeni eliphezulu, ukuqinisa nokupholisa izikhungo zedatha, nokubonisa ukuthi amamodeli ayathuthuka ngezindlela ezibalulekile. Ukuhlola kungaba yisithiyo ngokwako lapho ama-benchmarks egcwele noma ehluleka ukubonisa inani langempela. Emigwaqweni eminingi, imikhawulo yokuthunyelwa, izidingo zokuthembeka, kanye nomkhawulo wezindleko nakho kuyakha lokho okuhlala kuwusizo.
Kungani kubikwa ukuthi i-OpenAI ithuthukisa idivayisi ye-AI yesitayela sesipikha ehlakaniphile enekhamera?
Umbiko uphakamisa ukuthi i-OpenAI ifuna isiteshi sehadiwe yabathengi ngqo, kunokuba ibe ngaphakathi kwepulatifomu yomunye umuntu. Isipikha esihlakaniphile esinekhamera sikhomba usizo "oluzungezile" olungakwazi ukuhumusha okungaphezu kwezwi kuphela. Kubasebenzisi, izinto ezibalulekile okufanele ziqashelwe ubumfihlo, lapho ividiyo icutshungulwa khona, kanye nokuthi yiziphi izilawuli ezikhona zokukhubaza noma zokunciphisa ukuzwa ezindaweni zansuku zonke.
Kungani iNvidia ingatshala imali enkulu kangaka emzuliswaneni olandelayo wokuxhaswa ngezimali we-OpenAI?
Intshisekelo ebikiwe ibonisa umjikelezo weqhinga: I-Nvidia iyazuza lapho isidingo se-AI sikhuphuka, futhi isabelo esijulile kumakhi omkhulu wemodeli singaqinisa lokho kuhambisana. Kuphinde kubonise ukuthi ukunikezwa kwekhompyutha okuhlangene kanye nokuthuthukiswa kwemodeli yemingcele sekuhlangene kangakanani. Uma lokhu kutshalwa kwezimali kwenzeka ngezinga okwakuhletshwa ngalo, kuzogcizelela ukuthi imali nengqalasizinda ingakanani manje ebalulekile ekuqhubekeni nokuncintisana ku-AI ephezulu.
Kuyini i-Anthropic's Claude Code Security, futhi ukuskena ikhodi okusebenzisa i-AI kusiza kanjani?
I-Claude Code Security ibekwe njengokutholwa kobuthakathaka okusizwa yi-AI okusikisela ukulungiswa kwabantu okufanele bakubukeze. Ukugcizelela "komuntu ngaphakathi kwe-loop" kubalulekile ngoba ama-patches azenzakalelayo angaletha ukuhlehla noma izinguquko ezingaphephile uma zisetshenziswa ngokungacabangi. Empeleni, la mathuluzi angasheshisa i-triage, aveze amaphethini ayingozi kusenesikhathi, futhi anciphise isikhathi sokulungisa - ikakhulukazi lapho abavikeli bedinga ukuhambisana nokutholwa okusheshayo okunikwe amandla yi-AI ohlangothini lomhlaseli.
Ingabe izinguquko zobuholi bemidlalo ye-Microsoft zizoholela kokuqukethwe okwengeziwe kwe-AI, noma "ukungalungi kwe-AI" okuncane?
Lo myalezo usikisela ukuthi iMicrosoft ifuna ukugwema ukugcwala imidlalo nge-“AI slop” esezingeni eliphansi, ekhiqizwe ngobuningi, ngisho noma i-AI iba yingxenye yengxoxo yesu. Ubuholi obuvela enhlanganweni eseduze ne-AI akusho ngokuzenzakalelayo okuqukethwe okungapheli okukhiqizwe, kodwa kusho ukuzama okwengeziwe kumathuluzi, ukwenza kube ngokwakho, kanye nemisebenzi yokukhiqiza. Ukuhlolwa kwangempela kuzoba ukuthi ngabe i-guardrails yekhwalithi iyasebenza yini uma ukwenza ngokuzenzakalelayo kwenza okuqukethwe kushibhile ukukukhiqiza.