Umbhali: Ulink Media
Ukudweba i-AI akuqedanga ukushisa, imibuzo nezimpendulo ze-AI futhi kwaqala ubuhlakani obusha!
Ungakukholwa? Ikhono lokukhiqiza ikhodi ngqo, ukulungisa amaphutha ngokuzenzakalelayo, ukwenza izingxoxo ku-inthanethi, ukubhala izikripthi zesimo, izinkondlo, amanoveli, ngisho nokubhala izinhlelo zokubhubhisa abantu… Lokhu kuvela ku-chatbot esekelwe ku-AI.
Ngomhlaka-30 Novemba, i-OpenAI yethule uhlelo lwezingxoxo olusekelwe ku-AI olubizwa nge-ChatGPT, i-chatbot. Ngokusho kwezikhulu, i-ChatGPT iyakwazi ukuxhumana ngendlela yengxoxo, futhi ifomethi yengxoxo ivumela i-ChatGPT ukuphendula imibuzo yokulandelela, ukuvuma amaphutha, ukuphikisa imibono engalungile, nokwenqaba izicelo ezingafanele.
Ngokusho kwedatha, i-OpenAI yasungulwa ngo-2015. Iyinkampani yocwaningo lwezobunhloli bokwenziwa eyasungulwa ngokubambisana nguMusk, uSam Altman nabanye. Ihlose ukufeza ubuhlakani be-General Artificial (AGI) obuvikelekile futhi yethule ubuchwepheshe bezobunhloli bokwenziwa obuhlanganisa i-Dactyl, i-GFT-2 kanye ne-DALL-E.
Kodwa-ke, i-ChatGPT iwuhlobo oluvela kumodeli ye-GPT-3, okwamanje oluku-beta futhi lumahhala kulabo abane-akhawunti ye-OpenAI, kodwa imodeli yenkampani ezayo ye-GPT-4 izoba namandla kakhulu.
I-spin-off eyodwa, eseseku-beta yamahhala, isivele idonsele abasebenzisi abangaphezu kwesigidi, lapho uMusk ebhala ku-Twitter ethi: I-ChatGPT iyesabeka futhi siseduze ne-AI eyingozi nenamandla. Ngakho-ke, wake wazibuza ukuthi i-ChatGPT imayelana nani? Yalethani?
Kungani i-ChatGPT ithandwa kangaka kwi-inthanethi?
Ngokuphathelene nentuthuko, i-ChatGPT ihlelwe kahle kusukela kumodeli emndenini we-GPT-3.5, kanti i-ChatGPT kanye ne-GPT-3.5 ziqeqeshwe ngengqalasizinda ye-Azure AI supercomputing. Futhi, i-ChatGPT iyilungu le-InstructGPT, eliqeqeshwa yi-InstructGPT ngendlela efanayo ethi “Reinforcement Learning from Human Feedback (RLHF)”, kodwa ngezilungiselelo zokuqoqwa kwedatha ezihlukile kancane.
I-ChatGPT esekelwe ekuqeqeshweni kwe-RLHF, njengemodeli yolimi oluxoxisanayo, ingalingisa ukuziphatha komuntu ukuqhuba ingxoxo yolimi lwemvelo eqhubekayo.
Uma uxhumana nabasebenzisi, i-ChatGPT ingahlola ngokugcwele izidingo zangempela zabasebenzisi futhi inikeze izimpendulo abazidingayo noma ngabe abasebenzisi abakwazi ukuchaza imibuzo ngokunembile. Futhi okuqukethwe kwempendulo ukumboza izilinganiso eziningi, ikhwalithi yokuqukethwe ayingaphansi kwe-"injini yokusesha" ye-Google, ngokusebenza kwayo inamandla kune-Google, ngoba le ngxenye yomsebenzisi ithumele umuzwa othi: "I-Google ilahliwe!"
Ngaphezu kwalokho, i-ChatGPT ingakusiza ukuthi ubhale izinhlelo ezikhiqiza ikhodi ngqo. I-ChatGPT inezisekelo zokuhlela. Ayigcini nje ngokuhlinzeka ngekhodi ongayisebenzisa, kodwa futhi ibhala nemibono yokusebenzisa. I-ChatGPT ingathola futhi amaphutha kukhodi yakho futhi inikeze izincazelo ezinemininingwane zokuthi yini engahambanga kahle nokuthi ungayilungisa kanjani.
Vele, uma i-ChatGPT ingabamba izinhliziyo zezigidi zabasebenzisi ngalezi zici ezimbili kuphela, unephutha. I-ChatGPT inganikeza nezinkulumo, ibhale amaphepha, ibhale amanoveli, yenze izingxoxo ze-AI online, iklame amakamelo okulala, njalo njalo.
Ngakho-ke akunangqondo ukuthi i-ChatGPT ixhume izigidi zabasebenzisi ngezimo zayo ezahlukahlukene ze-AI. Kodwa empeleni, i-ChatGPT iqeqeshwe ngabantu, futhi yize ihlakaniphile, ingenza amaphutha. Isenamaphutha athile ekhonweni lolimi, futhi ukuthembeka kwezimpendulo zayo kusazocatshangelwa. Vele, okwamanje, i-OpenAI ivulelekile futhi mayelana nokulinganiselwa kwe-ChatGPT.
USam Altman, i-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, kokubili umdwebo we-AI owasabalala esikhathini esithile esidlule kanye ne-ChatGPT eyaheha abantu abaningi be-inthanethi kusobala ukuthi kukhomba esihlokweni esisodwa - i-AIGC. Okubizwa ngokuthi i-AIGC, Okuqukethwe Okukhiqizwe yi-AI, kubhekisela esizukulwaneni esisha sokuqukethwe Okukhiqizwe ngokuzenzakalelayo ubuchwepheshe be-AI ngemuva kwe-UGC kanye ne-PGC.
Ngakho-ke, akunzima ukuthola ukuthi esinye sezizathu eziyinhloko zokuthandwa komdwebo we-AI ukuthi imodeli yomdwebo we-AI ingaqonda ngqo okokufaka kolimi komsebenzisi, futhi ihlanganise ngokuseduze ukuqonda okuqukethwe kolimi kanye nokuqonda okuqukethwe kwesithombe kumodeli. I-ChatGPT iphinde yathola ukunakwa njengemodeli yolimi lwemvelo esebenzisana nayo.
Akunakuphikwa ukuthi ngentuthuko esheshayo yobuhlakani bokwenziwa eminyakeni yamuva nje, i-AIGC ingenisa igagasi elisha lezimo zohlelo lokusebenza. Ividiyo yezithombe ze-AI, ukudweba kwe-AI kanye neminye imisebenzi emele yenza isithombe se-AIGC singabonakala yonke indawo kuvidiyo emfushane, ukusakazwa bukhoma, ukubamba kanye nesiteji sephathi, okuqinisekisa ne-AIGC enamandla.
Ngokusho kukaGartner, i-AI ekhiqizayo izobalelwa ku-10% wayo yonke idatha ekhiqizwe ngo-2025. Ngaphezu kwalokho, uGuotai Junan uphinde wathi eminyakeni emihlanu ezayo, u-10%-30% wokuqukethwe kwesithombe kungase kukhiqizwe yi-AI, futhi usayizi wemakethe ohambisanayo ungadlula ama-yuan ayizigidigidi ezingama-60.
Kungabonakala ukuthi i-AIGC isheshisa ukuhlanganiswa okujulile kanye nentuthuko kuzo zonke izindlela zokuphila, futhi amathuba ayo okuthuthukiswa abanzi kakhulu. Kodwa-ke, akunakuphikwa ukuthi kusenezingxabano eziningi enqubweni yokuthuthukiswa kwe-AIGC. Uchungechunge lwezimboni aluphelele, ubuchwepheshe abukavuthwa ngokwanele, izinkinga zobunikazi be-copyright nokunye, ikakhulukazi ngenkinga "yokufaka esikhundleni se-AI yomuntu", ngezinga elithile, ukuthuthukiswa kwe-AIGC kuyavinjelwa. Kodwa-ke, uXiaobian ukholelwa ukuthi i-AIGC ingangena embonweni womphakathi, futhi iguqule izimo zokusetshenziswa kwezimboni eziningi, kumele ibe nezimfanelo zayo, futhi amandla ayo okuthuthukiswa adinga ukuthuthukiswa kabanzi.
Isikhathi sokuthunyelwe: Disemba 12-2022