Kwenzekeni
I-Seoul Economic Daily ibike ukuthi i-Alibaba isanda kubamba umhlangano wonjiniyela we-Qwen eSeoul, kuyilapho i-Nvidia kanye ne-OpenAI isingathe umcimbi wonjiniyela ekuqaleni kwale nyanga. Isitolo siphinde sabika ukuthi i-Anthropic inikeze amalungu e-Korea Startup Forum u-$10,000 kumakhredithi e-Claude, nezinkampani ezibalelwa ku-380 ezifake izicelo futhi ezicishe zibe ngu-190 zijoyine inkundla ngemuva kokuqala komnikelo.
I-Seoul Economic Daily ibike ngo-Aug. 26 ukuthi i-Alibaba isanda kubamba umcimbi we-Qwen Meetup Seoul esifundeni sase-Seoul's Gangnam yabathuthukisi bokuqalisa. Isitolo sithe u-Edwin Tack, umakhi wezixazululo e-Alibaba Cloud Korea, uxoxe ngezindlela zokulawula izindleko zalokho okubizwa nge-vibe coding ngemodeli ye-Alibaba's Qwen3.8-Max. I-Seoul Economic Daily ibike ukuthi i-Alibaba ichaza imodeli njengenamapharamitha ayizigidi eziyizigidi eziyizinkulungwane ezingu-2.4, inani eliphumayo uma liqhathaniswa nesilinganiso samamodeli ambalwa asemngceleni athuthukiswe e-United States. Izibalo zepharamitha zizodwa azisunguli ikhwalithi yemodeli, ukusebenza kahle noma ukufaneleka kwebhizinisi elithile. Lo mcimbi ubuye wabhekana nezindlela zokusebenzisa i-AI ngenkathi ulawula idatha yenkampani futhi unciphisa ubungozi bolwazi olushiya inkampani, ngokusho kombiko.
Isitolo sibike ukuthi i-Nvidia ne-OpenAI ziphinde zandisa ukuxhumana okuqondile nabathuthukisi bokuqalisa base-Korea. I-Seoul Economic Daily ithe i-Nvidia ibambe i-Korea Developer Meetup ehhovisi le-OpenAI eSeoul ngo-Aug. 12. Lo mcimbi ubugubha iminyaka engu-10 ye-Nvidia ukulethwa kwe-DGX-1 i-deep-learning supercomputer ku-ZXZAIU0 wathi uhlelo lugxile kakhulu ku-ZXZAIU0. Amathuluzi e-AI-ejenti ye-Nvidia kanye ne-OpenAI's Codex-based-software-development workflow. I-athikili ayicacisi ukuthi yimaphi amathuluzi abonisiwe, zingaki iziqalo ezithanyelwe, noma ngabe ukufinyelela kunikezwe ababambiqhaza noma ukuthi ngabe yiziphi izinkampani ezathatha imikhiqizo kamuva.
I-Seoul Economic Daily ibike ngokwehlukana ukuthi i-Anthropic isekele imodeli yayo ye-Claude nge-Korea Startup Forum, echazwe esihlokweni njengenhlangano enkulu yokuqalisa yaseNingizimu Korea, kusukela ngoMashi. Ngaphansi kohlelo, izinkampani ezingamalungu zithola amakhredithi entengo engu-$10,000 we-Claude, ngokusho kwesitolo. Cishe izinkampani ezingamalungu angama-380 zazifake izicelo, futhi cishe eziyi-190 kwakuyiziqalo ezijoyine inkundla ngemuva kokuqala kokwesekwa kwe-Anthropic. Umjikelezo wokuqala wokufaka izicelo zesithangami kubikwa ukuthi uvalwe ngokushesha ngemuva kokwanda kwemibuzo yobulungu, kanti le nhlangano ihlela ukuvula umzuliswano wesibili. I-athikili ayisho ukuthi amakhredithi amelela uxhaso olulingana nemali, ukufinyelela kokuphromotha okukhawulelwe ukusetshenziswa noma olunye uhlobo loxhaso.
Imininingwane yomthombo: en.sedaily.com ↗
Kungani kubalulekile
Umsebenzi obikiwe ubonisa ukuthi izinkampani ze-AI ziqhudelana kanjani nonjiniyela nabaqalayo ngokuhlanganisa ukufinyelela okuyimodeli, amathuluzi esoftware, isiqondiso sobuchwepheshe kanye nokusetshenziswa okuxhasiwe. Ukutholwa kusenesikhathi kungase kuthonye ukuthi iyiphi imodeli yezinhlelo zemvelo izinkampani zase-Korea ezakha emikhiqizweni yazo, nakuba i-athikili ingaqinisekisi ngokuzimele ukuthi le mizamo ikhiqize ukuthunyelwa okukhulu noma imali engenayo.
Izinhlelo ezibikiwe zibalulekile ngoba abaqalayo bavame ukwenza ukukhetha kobuchwepheshe obuyisisekelo ngaphambi kokuba babe nezinsiza zokwakha yonke ingxenye yesitaki se-AI ngokwabo. Ukufinyelela kumamodeli amakhredithi kunganciphisa izindleko zokuhlola kusenesikhathi, kuyilapho imicimbi yezobuchwepheshe ingasiza onjiniyela bakhethe amathuluzi, balungiselele izinhlelo zokusebenza futhi banqume ukuthi idatha nomsebenzi uzosebenza kuphi. Umbiko we-Seoul Economic Daily wethula lo msebenzi njengomzamo ohlanganisiwe wezinkampani ezinkulu ze-AI ukuheha ama-startups aseKorea kusenesikhathi, kunokuba kube nje ukuboniswa kwemikhiqizo ehlukene.
Ababukeli bemboni yale ndatshana bathi ama-startups aqondiswe njengebhuloho emakethe ebanzi yaseKorea. Ukucabanga kwabo, njengoba kubikwe yi-Seoul Economic Daily, ukuthi i-ecosystem ihlanganisa okungaphezu kwe-chip noma imodeli: Inkundla ye-Nvidia, isibonelo, ihlanganisa imitapo yolwazi esekwe ku-CUDA, abahlanganisi namathuluzi okuthuthukisa. Uma abathuthukisi sebethole amakhono nezinkampani zakha izinqubo zokusebenza ezizungeze lezo zingxenye, ukushintshela endaweni eyimbangi kungase kudinge izinguquko kusofthiwe, ukuqeqeshwa nengqalasizinda. Lokho ukuhlola komkhakha okubikwe yisikhungo, hhayi ukutholwa kukalwa ngokuzimela mayelana neziqalo zase-Korea.
Okokuqala, ukuncintisana phakathi kwabahlinzeki bomhlaba kungaletha amanani aphansi, ukusekelwa okwengeziwe kobuchwepheshe nokukhetha okukhulu. Lo mbhalo ucaphune isikhulu semboni esingadalulwanga sithi ukuvela kwamamodeli aseShayina anekhono kube nomthelela ekwehliseni amanani entengo kumamodeli ayo akamuva OpenAI. Ngenxa yokuthi lowo mthombo awaziwa futhi ingcaphuno yahunyushwa isuka ku-Korean, isimangalo namagama akuqinisekiswanga ngokuzimela lapha. Umthelela osebenzayo uzoncika ekutheni izaphulelo namakhredithi azohlala etholakala ngemva kokuba izinkampani zakha imikhiqizo yazo, noma ngabe abahlinzeki babeka idatha noma imikhawulo yokusetshenziswa nokuthi ukuqalisa kungahamba yini phakathi kwamamodeli ngaphandle kwezindleko ezinkulu zobuchwepheshe.
Lo mbiko uphakamisa nombuzo wokuphathwa kwedatha. Umhlangano ka-Alibaba kubikwa ukuthi uhlanganise izindlela zokuvimbela imininingwane yebhizinisi ukuthi ingavuzi ngaphandle kwenkampani, okusikisela ukuthi ukuphathwa kwedatha kuyingxenye yesinqumo sokutholwa. Umthombo awunikezi incazelo yobuchwepheshe yalezo zilawuli, imiphumela yocwaningo noma ukuvikela kwenkontileka. Izinkampani ezicabangela lezi zinhlelo kuzodingeka zihlole ukugcinwa, ukusetshenziswa kokuqeqeshwa, izimvume zokufinyelela, ukucutshungulwa kwendawo, impendulo yezehlakalo kanye nemibandela yokuphuma esikhundleni sokuphatha amakhredithi amahhala noma ukwesekwa kobuchwepheshe njengobufakazi bokuphepha.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
What is a common training objective for an autoregressive language model?
Ongakubuka ngokulandelayo
Okungaziwa okubalulekile ukuthi izehlakalo ezibikiwe ziholela ekusetshenzisweni okuqhubekayo, okuyiziqalo ezithola amakhredithi, imigomo eqondile yezinhlelo nokuthi ukucaciswa kwemodeli okubikiwe kwe-Alibaba kungaqinisekiswa ngokuzimele. Ukubika okwengeziwe kufanele futhi kuqinisekise ukuthi le mizamo inweba ngaphezu kokufinyelela konjiniyela ibe ubambiswano lwezentengiso noma ukwamukelwa okubanzi kwemakethe yase-Korea.
Okubalulekile kokuqala ukuqinisekiswa okuzimele kwemigomo nesilinganiso sezinhlelo. I-Alibaba, Nvidia, OpenAI kanye ne-Anthropic zingacacisa ukuthi imihlangano ebikiwe bekuyimicimbi yonjiniyela evuliwe, amaseshini ezimemo kuphela noma izinhlelo zokuqalisa ezisemthethweni. Okunikezwayo kwe-Anthropic, imininingwane ebalulekile engekho ihlanganisa ukufaneleka, izinsuku zokuphelelwa yisikhathi, imikhawulo yokusetshenziswa, ukufakwa kwamamodeli, amalungelo edatha kanye nokuthi izicelo ezibikiwe ezingu-380 kanye nobulungu bangemva kohlelo abayi-190 babalwe kanjani. I-Seoul Economic Daily ingumthombo oqanjwe walezo zibalo; aziqinisekiswanga ngokuzimela ezintweni ezinikeziwe.
Ukubika kokulandelela kufanele kubheke ubufakazi bemiphumela kunokuhambela. Izinkomba eziwusizo zingabandakanya ukuqalisa ukuhambisa izinhlelo zokusebenza emkhiqizweni, ubambiswano lwezentengiselwano olusayiniwe, izinguquko ezilinganisekayo ezindlekweni zokucabanga, ukusetshenziswa okuqhubekayo ngemva kokuphelelwa yisikhathi kwamakhredithi nezibonelo zemikhiqizo yaseKorea eyakhiwe ngamathuluzi ezinkampani ezibamba iqhaza. Umthombo awukhombi iziqalo ezibamba iqhaza noma ukuthunyelwa kwemibiko, imali engenayo, ukuhlolwa kokusebenza noma imiphumela yekhasimende. Ngaphandle kwaleyo mininingwane, indaba isungula umzamo wokuqomisana osebenzayo kodwa hhayi ukushintshwa okubonisiwe kusabelo semakethe.
Isibalo esibikiwe esingu-2.4-trillion- se-Qwen3.8-Max siphinde sigunyaze ukuqinisekiswa. Isibalo sepharamitha singaba nzima ukusitolika kuzo zonke izakhiwo futhi asibonisi ngokwaso ukuthi imodeli isebenza kanjani, kubiza malini ukusebenza noma ukuthi iphatha kahle imisebenzi yolimi lwesi-Korean neyebhizinisi. Ukuhlola okuzimele kufanele kuhlole ukunemba, ukubambezeleka, ukwethembeka, ukulayisensa, ubumfihlo kanye nengqikithi yezindleko zokusebenza. Ukuhlolisisa okufanayo kufanele kusebenze ezimangalweni ezimayelana namathuluzi e-AI-ejenti kanye nokugeleza komsebenzi okusekelwe ku-Codex, lapho i-athikili inganikezi imiphumela yokuqhathanisa.
Okokugcina, bheka ukuthi lezi zinhlelo ezibheke ekuqaleni ziyanwebeka yini zibe izivumelwano zikahulumeni, zamabhizinisi amakhulu noma zengqalasizinda eSouth Korea. Ukuhlangana konjiniyela noma uhlelo lwekhredithi lungahlala luwumzamo olinganiselwe wokumaketha nokugibela, noma lungaba yingxenye yomncintiswano obanzi mayelana nezinkontileka zamafu, ukuthunyelwa kwama-chip namandla kazwelonke e-AI. Umbiko onikeziwe awuqinisekisi ukuthi iyiphi indlela evelayo, futhi awukho umcimbi oqhubekayo onembile kungobo yomlando wangaphakathi ofanelekayo ofana nalo msebenzi wokuxhumana ohlanganisiwe.