Umbhali: Ulink Media
Umdwebo we-AI awukaqedi ukushisa, i-AI Q&A futhi uqalise ukuhlanya okusha!
Ungakukholwa? Ikhono lokukhiqiza ikhodi ngokuqondile, ukulungisa iziphazamisi ngokuzenzakalelayo, ukubonisana nge-inthanethi, ukubhala imibhalo yesimo, izinkondlo, amanoveli, ngisho nokubhala izinhlelo zokucekela phansi abantu... Lokhu kuphuma ku-chatbot esekelwe ku-AI.
Ngomhla zingama-30 kuNovemba, i-OpenAI yethula uhlelo lwezingxoxo olusekelwe ku-AI olubizwa nge-ChatGPT, i-chatbot. Ngokusho kwezikhulu, i-ChatGPT iyakwazi ukuxhumana ngendlela yengxoxo, futhi ifomethi yengxoxo yenza i-ChatGPT ikwazi ukuphendula imibuzo yokulandelela, ivume amaphutha, inselele izakhiwo ezingalungile, futhi yenqabe izicelo ezingafanele.
Ngokusho kwedatha, i-OpenAI yasungulwa ku-2015. Kuyinkampani yocwaningo lwezobunhloli yokwenziwa eyasungulwa nguMusk, uSam Altman nabanye. Ihlose ukuzuza ubuhlakani obujwayelekile be-Artificial Artificial (AGI) obuvikelekile futhi yethule ubuchwepheshe bobuhlakani bokwenziwa obuhlanganisa i-Dactyl, i-GFT-2 ne-DALL-E.
Nokho, i-ChatGPT iphuma kuphela kumodeli ye-GPT-3, okwamanje eku-beta futhi imahhala kulabo abane-akhawunti ye-OpenAI, kodwa imodeli yenkampani ezayo ye-GPT-4 izoba namandla nakakhulu.
I-spin-off eyodwa, okusese ku-beta yamahhala, isivele ihehe abasebenzisi abangaphezu kwesigidi, nge-Musk tweeting: I-ChatGPT iyesabeka futhi siseduze ne-AI eyingozi nenamandla. Ngakho-ke, wake wazibuza ukuthi i-ChatGPT imayelana nani? Lalethani?
Kungani i-ChatGPT idume kangaka ku-inthanethi?
Ngokuqondene nentuthuko, i-ChatGPT icushwe kahle isuka kumodeli emndenini we-GPT-3.5, futhi i-ChatGPT ne-GPT-3.5 baqeqeshwe kwingqalasizinda ye-Azure AI ye-supercomputing. Futhi, i-ChatGPT iyizelamani ku-InstructGPT, i-InstructGPT eyiqeqesha ngendlela efanayo “Yokuqiniswa Ukufunda Ngempendulo Yomuntu (RLHF)”, kodwa ngezilungiselelo zokuqoqwa kwedatha ezihluke kancane.
I-ChatGPT esekelwe ekuqeqeshweni kwe-RLHF, njengemodeli yolimi lwengxoxo, ingalingisa ukuziphatha komuntu ukuze iqhube ingxoxo yolimi lwemvelo eqhubekayo.
Lapho usebenzisana nabasebenzisi, i-ChatGPT ingahlola ngokugcwele izidingo zangempela zabasebenzisi futhi inikeze izimpendulo abazidingayo ngisho noma abasebenzisi bengakwazi ukuchaza ngokunembile imibuzo. Futhi okuqukethwe kwempendulo ukumboza izilinganiso eziningi, ikhwalithi yokuqukethwe ayikho ngaphansi “kwenjini yosesho” ye-Google, ngokungenzeka iqine kune-Google, ngoba le ngxenye yomsebenzisi ithumele umuzwa: “I-Google ilahlekile!
Ngaphezu kwalokho, i-ChatGPT ingakusiza ukuthi ubhale izinhlelo ezikhiqiza ikhodi ngokuqondile. I-ChatGPT inezisekelo zokuhlela. Ayinikezi kuphela ikhodi okufanele isetshenziswe, kodwa futhi ibhala imibono yokuqaliswa. I-ChatGPT ingathola futhi iziphazamisi kukhodi yakho futhi ikunikeze izincazelo ezinemininingwane yalokho okungahambanga kahle nokuthi zingalungiswa kanjani.
Impela, uma i-ChatGPT ingathwebula izinhliziyo zezigidi zabasebenzisi ngalezi zici ezimbili nje, unephutha. I-ChatGPT inganikeza nezinkulumo, ibhale amaphepha, ibhale amanoveli, yenze ukubonisana nge-AI ku-inthanethi, idizayine amagumbi okulala, nokunye.
Ngakho-ke akukhona okungenangqondo ukuthi i-ChatGPT ixhume izigidi zabasebenzisi ngezimo zayo ezihlukahlukene ze-AI. Kodwa empeleni, i-ChatGPT iqeqeshwa abantu, futhi nakuba ihlakaniphile, iyakwazi ukwenza amaphutha. Isenakho ukushiyeka ekhonweni lolimi, futhi ukwethembeka kwezimpendulo zayo kusazobhekwa. Vele, kuleli phuzu, i-OpenAI nayo ivuliwe mayelana nemikhawulo ye-ChatGPT.
USam Altman, oyi-CEO ye-OpenAI, uthe ukuxhumana kwezilimi kuyikusasa, nokuthi i-ChatGPT iyisibonelo sokuqala sekusasa lapho abasizi be-AI bengaxoxa nabasebenzisi, baphendule imibuzo, futhi banikeze iziphakamiso.
Kuze kube nini i-AIGC ifika?
Eqinisweni, yomibili imidwebo ye-AI eyaba negciwane esikhathini esidlule kanye ne-ChatGPT eyaheha inqwaba yabasebenzisi benethiwekhi ikhomba ngokusobala esihlokweni esisodwa - i-AIGC. Okubizwa nge-AIGC, Okuqukethwe okukhiqizwe yi-AI, kubhekisela esizukulwaneni esisha sokuqukethwe Okukhiqizwa ngokuzenzakalela ubuchwepheshe be-AI ngemva kwe-UGC ne-PGC.
Ngakho-ke, akunzima ukuthola ukuthi esinye sezizathu eziyinhloko zokuthandwa komdwebo we-AI ukuthi imodeli yokudweba ye-AI ingakwazi ukuqonda ngokuqondile okokufaka kolimi lomsebenzisi, futhi ihlanganise eduze ukuqonda kokuqukethwe kolimi nokuqonda kokuqukethwe kwesithombe kumodeli. I-ChatGPT iphinde yathola ukunakwa njengemodeli yolimi lwemvelo esebenzisanayo.
Ngokungangabazeki, ngokuthuthuka okusheshayo kobuhlakani bokwenziwa eminyakeni yamuva nje, i-AIGC ingenisa igagasi elisha lezimo zohlelo lokusebenza. Ividiyo eyingcaca ye-AI, imidwebo ye-AI kanye neminye imisebenzi emele yenza isibalo se-AIGC sibonakale yonke indawo kuvidiyo emfushane, ukusakazwa bukhoma, ukusingathwa kanye nesiteji sephathi, nakho okuqinisekisa i-AIGC enamandla.
Ngokusho kukaGartner, i-AI yokukhiqiza izobalelwa ku-10% wayo yonke idatha ekhiqizwe ngo-2025. Ngaphezu kwalokho, uGuotai Junan uphinde wathi eminyakeni emihlanu ezayo, i-10% -30% yokuqukethwe kwesithombe ingakhiqizwa yi-AI, futhi okuhambisanayo. usayizi wemakethe ungadlula ama-yuan ayizigidi eziyizinkulungwane ezingama-60.
Kungabonakala ukuthi i-AIGC isheshisa ukuhlanganiswa nokuthuthuka okujulile ngayo yonke imikhakha yempilo, futhi ithemba layo lokuthuthuka libanzi kakhulu. Nokho, akunakuphikwa ukuthi kusenezingxabano eziningi ohlelweni lokuthuthukiswa kwe-AIGC. Uchungechunge lwezimboni aluphelele, ubuchwepheshe abukavuthwa ngokwanele, izinkinga zobunikazi be-copyright nokunye, ikakhulukazi mayelana nenkinga "ye-AI esikhundleni somuntu", ngezinga elithile, ukuthuthukiswa kwe-AIGC kuyavinjelwa. Kodwa-ke, u-Xiaobian ukholelwa ukuthi i-AIGC ingangena embonweni womphakathi, futhi ilungise kabusha izimo zezicelo zezimboni eziningi, kufanele ibe nokufaneleka kwayo, futhi amandla ayo okuthuthuka adinga ukuthuthukiswa.
Isikhathi sokuthumela: Dec-12-2022