Umbhali: Ulink Media
Ipeyinti ye-AI ayibuchithanga ubushushu, i-AI Q&A kwaye yaqalisa i-craze entsha!
Ngaba unokukholelwa? Ikhono lokuvelisa ngokuthe ngqo ikhowudi, ukulungisa ngokuzenzekelayo iimpazamo, ukwenza ukubonisana kwi-intanethi, ukubhala izikripthi zeemeko, imibongo, iinoveli, kunye nokubhala izicwangciso zokutshabalalisa abantu ... Ezi ziphuma kwi-AI-based chatbot.
Nge-30 kaNovemba, i-OpenAI yasungula inkqubo yencoko esekwe kwi-AI ebizwa ngokuba yi-ChatGPT, i-chatbot. Ngokwamagosa, i-ChatGPT iyakwazi ukusebenzisana ngendlela yencoko, kwaye ifomathi yencoko yenza ukuba i-ChatGPT iphendule imibuzo elandelayo, ivume iimpazamo, icele umngeni kwizakhiwo ezingachanekanga, kwaye yale izicelo ezingafanelekanga.
Ngokutsho kwedatha, i-OpenAI yasungulwa kwi-2015. Yinkampani yophando lwengqondo eyenziwe ngokubambisana nguMusk, uSam Altman nabanye. Ijolise ekufumaneni ukhuseleko lobukrelekrele Jikelele beArtificial (AGI) kwaye yazise ubugcisa bobuchule bokwenziwa obuquka iDactyl, GFT-2 kunye neDALL-E.
Nangona kunjalo, i-ChatGPT iphuma kuphela kwimodeli ye-GPT-3, okwangoku ikwi-beta kwaye isimahla kwabo bane-akhawunti ye-OpenAI, kodwa imodeli ye-GPT-4 yenkampani ezayo iya kuba namandla ngakumbi.
I-spin-off enye, esekho kwi-beta yamahhala, sele itsale abasebenzisi abangaphezu kwesigidi, kunye ne-Musk tweeting: I-ChatGPT iyoyikisa kwaye sisondele kwi-AI eyingozi kwaye inamandla. Ke, ngaba ukhe wazibuza ukuba i-ChatGPT imalunga nantoni? Izise ntoni?
Kutheni i-ChatGPT ithandwa kangaka kwi-Intanethi?
Ngokubhekiselele kuphuhliso, i-ChatGPT ilungiswe kakuhle ukusuka kwimodeli kwintsapho ye-GPT-3.5, kwaye i-ChatGPT kunye ne-GPT-3.5 baqeqeshwe kwi-Azure AI supercomputing infrastructure. Kwakhona, i-ChatGPT ngumntakwe-InstructGPT, leyo i-InstructGPT iqeqesha ngendlela efanayo “yokuFundisa ngokuFundisa kwiNgxelo yoMntu (RLHF)”, kodwa ngoSeto lokuqokelela idatha olwahluke kancinane.
I-ChatGPT esekelwe kuqeqesho lwe-RLHF, njengemodeli yolwimi lwencoko, inokuxelisa ukuziphatha komntu ukuqhuba ingxoxo eqhubekayo yolwimi lwendalo.
Xa usebenzisana nabasebenzisi, i-ChatGPT inokuhlola ngokupheleleyo iimfuno zangempela zabasebenzisi kwaye inike iimpendulo abazifunayo nokuba abasebenzisi abanakukwazi ukuchaza ngokuchanekileyo imibuzo. Kwaye umxholo wempendulo ukugubungela imilinganiselo emininzi, umgangatho womxholo awukho ngaphantsi kwe "injini yokukhangela" kaGoogle, enokwenzeka yomelele kunoGoogle, kuba le nxalenye yomsebenzisi ithumele imvakalelo: "UGoogle ugwetyiwe!
Ukongeza, i-ChatGPT inokukunceda ukuba ubhale iinkqubo ezivelisa ikhowudi ngokuthe ngqo. I-ChatGPT ineziseko zocwangciso. Ayinikezeli kuphela ikhowudi yokusetyenziswa, kodwa ibhala neengcamango zokuphunyezwa. I-ChatGPT inokufumana ii-bugs kwikhowudi yakho kwaye inike iinkcazo ezineenkcukacha zokuba yintoni engalunganga kunye nendlela yokuyilungisa.
Ewe kunjalo, ukuba i-ChatGPT inokubamba iintliziyo zezigidi zabasebenzisi ngezi zinto zimbini nje, awulunganga. I-ChatGPT inokunika iintetho, ibhale amaphepha, ibhale iinoveli, yenza uthethwano lwe-AI kwi-intanethi, uyilo lwegumbi lokulala, njalo njalo.
Ke akukho ngqiqweni ukuba i-ChatGPT idibanise izigidi zabasebenzisi ngeendlela ezahlukeneyo ze-AI. Kodwa eneneni, i-ChatGPT iqeqeshwa ngabantu, kwaye nangona ikrelekrele, inokwenza iimpazamo. Isenokusilela kubuchule bolwimi, kwaye ukuthembeka kweempendulo zayo kusaza kuqwalaselwa. Ewe kunjalo, okwangoku, i-OpenAI ivulekile malunga nemida ye-ChatGPT.
USam Altman, i-CEO ye-OpenAI, uthe i-interfaces yolwimi yikamva, kwaye i-ChatGPT ngumzekelo wokuqala wexesha elizayo apho abancedisi be-AI banokuxoxa nabasebenzisi, baphendule imibuzo, kwaye banike iingcebiso.
Kude kube nini de i-AIGC ifike?
Ngapha koko, zombini umzobo we-AI oye waya kwintsholongwane kwixesha elidlulileyo kwaye i-ChatGPT eyatsala abantu abaninzi kwi-intanethi ikhomba ngokucacileyo kwisihloko esinye - i-AIGC. Into ebizwa ngokuba yi-AIGC, i-AI-generated Content, ibhekisela kwisizukulwana esitsha somxholo Eveliswa ngokuzenzekelayo ngeteknoloji ye-AI emva kwe-UGC kunye ne-PGC.
Ngako oko, akunzima ukufumanisa ukuba esinye sezizathu eziphambili zokuthandwa kwe-AI yokupenda kukuba imodeli yokudweba ye-AI inokuqonda ngokuthe ngqo igalelo lolwimi lomsebenzisi, kwaye idibanise ngokusondeleyo ukuqonda komxholo wolwimi kunye nokuqonda umxholo womfanekiso kwimodeli. I-ChatGPT ifumene ingqwalasela njengemodeli yolwimi lwendalo esebenzayo.
Ngokungathandabuzekiyo, ngophuhliso olukhawulezayo lobukrelekrele bokwenziwa kwiminyaka yamva nje, i-AIGC ingenisa amaza amatsha eemeko zokusetyenziswa. Ividiyo yegraphic ye-AI, umzobo we-AI kunye neminye imisebenzi emele yenza umzobo we-AIGC ubonakale kuyo yonke indawo kwividiyo emfutshane, ukusasazwa okubukhoma, ukusingathwa kunye neqonga leqela, elikwaqinisekisa i-AIGC enamandla.
Ngokutsho kukaGartner, i-AI yokuvelisa iya kuphendula i-10% yazo zonke iinkcukacha eziveliswayo ngo-2025. Ukongezelela, uGuotai Junan wathi kwakhona kwiminyaka emihlanu ezayo, i-10% -30% yomxholo womfanekiso unokuveliswa yi-AI, kunye neyodwa ehambelana nayo. ubungakanani bemarike bunokugqithisa i-yuan yebhiliyoni ezingama-60.
Ingabonwa ukuba i-AIGC ikhawulezisa ukudibanisa okunzulu kunye nophuhliso kunye nazo zonke iindlela zobomi, kwaye ithemba layo lophuhliso libanzi kakhulu. Nangona kunjalo, akunakuphikwa ukuba kusekho iingxabano ezininzi kwinkqubo yophuhliso lwe-AIGC. Ikhonkco yezoshishino ayifezekanga, iteknoloji ayikavuthwa ngokwaneleyo, imiba yobunini be-copyright kunye nokunye, ngakumbi malunga nengxaki "ye-AI yokutshintsha umntu", kwinqanaba elithile, ukuphuhliswa kwe-AIGC kuthintelwe. Nangona kunjalo, u-Xiaobian ukholelwa ukuba i-AIGC inokungena kumbono woluntu, kwaye ihlaziye kwakhona iimeko zesicelo kumashishini amaninzi, kufuneka ibe nemilinganiselo yayo, kwaye amandla ayo ophuhliso kufuneka aphuhliswe ngakumbi.
Ixesha lokuposa: Dec-12-2022