Umbhali: Ulink Media
Umzobo we-AI awuzange ubususe ubushushu, imibuzo neempendulo ze-AI kwaye uqalise inkanuko entsha!
Ungakholelwa? Ukukwazi ukuvelisa ikhowudi ngokuthe ngqo, ukulungisa iimpazamo ngokuzenzekelayo, ukwenza iingxoxo kwi-intanethi, ukubhala izikripthi zeemeko, imibongo, iinoveli, kwanokubhala izicwangciso zokutshabalalisa abantu... Ezi zivela kwi-chatbot esekwe kwi-AI.
Ngomhla wama-30 kweyeNkanga, i-OpenAI yaqalisa inkqubo yencoko esekwe kwi-AI ebizwa ngokuba yi-ChatGPT, i-chatbot. Ngokutsho kwamagosa, i-ChatGPT iyakwazi ukusebenzisana ngendlela yencoko, kwaye ifomathi yencoko ivumela i-ChatGPT ukuba iphendule imibuzo elandelayo, ivume iimpazamo, iphikise izizathu ezingalunganga, kwaye ikwale izicelo ezingafanelekanga.
Ngokwedatha, i-OpenAI yasekwa ngo-2015. Yinkampani yophando lobukrelekrele bokwenziwa eyasekwa nguMusk, uSam Altman nabanye. Ijolise ekufezekiseni ubukrelekrele bokwenziwa obukhuselekileyo (i-AGI) kwaye iqalise ubuchwepheshe bobukrelekrele bokwenziwa obuquka iDactyl, iGFT-2 kunye neDALL-E.
Nangona kunjalo, iChatGPT yinto evela kwimodeli yeGPT-3, okwangoku eku-beta kwaye isimahla kwabo baneakhawunti ye-OpenAI, kodwa imodeli yenkampani ezayo yeGPT-4 iya kuba namandla ngakumbi.
I-spin-off enye, esekwi-beta yasimahla, sele itsale abasebenzisi abangaphezu kwesigidi, uMusk ebhala kwi-tweet yakhe ethi: I-ChatGPT iyoyikisa kwaye sikufutshane ne-AI eyingozi nenamandla. Ngoko ke, ngaba wakha wazibuza ukuba i-ChatGPT imalunga nantoni? Izise ntoni?
Kutheni iChatGPT ithandwa kangaka kwi-intanethi?
Ngokuphathelele uphuhliso, iChatGPT ilungiswe kakuhle ukusuka kwimodeli yosapho lweGPT-3.5, kwaye iChatGPT kunye neGPT-3.5 ziqeqeshwe kwiziseko ze-Azure AI supercomputing. Kwakhona, iChatGPT yintsapho ye-InstructGPT, efundiswa yi-InstructGPT ngendlela efanayo "yokuQinisekisa uFundo oluvela kwiNgxelo yoLuntu (RLHF)", kodwa ngeSetingi zokuqokelela idatha ezahluke kancinci.
I-ChatGPT esekelwe kuqeqesho lwe-RLHF, njengomzekelo wolwimi oluthetha incoko, inokulinganisa ukuziphatha komntu ukuqhuba incoko yolwimi lwendalo eqhubekayo.
Xa usebenzisana nabasebenzisi, iChatGPT inokuhlola ngokupheleleyo iimfuno zokwenyani zabasebenzisi kwaye inike iimpendulo abazidingayo nokuba abasebenzisi abakwazi ukuchaza ngokuchanekileyo imibuzo. Kwaye umxholo wempendulo ugubungela iinkalo ezininzi, umgangatho womxholo awukho ngaphantsi kwe "injini yokukhangela" kaGoogle, ngokusebenzisekayo unamandla kuneGoogle, kuba eli candelo lomsebenzisi lithumele uluvo: "IGoogle iphelelwe lixesha!
Ukongeza, iChatGPT inokukunceda ubhale iinkqubo ezivelisa ikhowudi ngqo. IChatGPT ineziseko zokubhala iinkqubo. Ayiboneleli nje kuphela ngekhowudi oza kuyisebenzisa, kodwa ikwabhala neengcamango zokuphumeza. IChatGPT inokufumana iimpazamo kwikhowudi yakho kwaye inike iinkcazo ezineenkcukacha zezinto ezingahambanga kakuhle kunye nendlela yokuzilungisa.
Kakade ke, ukuba iChatGPT inokubamba iintliziyo zezigidi zabasebenzisi ngezi mpawu zimbini kuphela, uyaphazama. IChatGPT inokwenza iintetho, ibhale amaphepha, ibhale iinoveli, yenze iingxoxo ze-AI kwi-intanethi, iyile amagumbi okulala, njalo njalo.
Ngoko ke akukho ngqiqweni ukuba iChatGPT itsale izigidi zabasebenzisi ngeemeko zayo ezahlukeneyo ze-AI. Kodwa eneneni, iChatGPT iqeqeshwe ngabantu, kwaye nangona ikrelekrele, inokwenza iimpazamo. Iseneziphene ezithile kubuchule bolwimi, kwaye ukuthembeka kweempendulo zayo kusafuneka kuqwalaselwe. Kakade ke, okwangoku, i-OpenAI ikwavulelekile malunga nemida yeChatGPT.
USam Altman, i-CEO ye-OpenAI, uthe iindlela zokunxibelelana ngolwimi zizinto ezizayo, kwaye i-ChatGPT ngumzekelo wokuqala wexesha elizayo apho abancedisi be-AI banokuthetha nabasebenzisi, baphendule imibuzo, kwaye banike neengcebiso.
Kuza kuthatha ixesha elingakanani ngaphambi kokuba i-AIGC ifike?
Enyanisweni, zombini umzobo we-AI owasasazeka kwixesha elidlulileyo kunye ne-ChatGPT eyatsala abantu abaninzi be-intanethi ngokucacileyo zikhomba kwisihloko esinye - i-AIGC. Okubizwa ngokuba yi-AIGC, umxholo oveliswe yi-AI, kubhekisa kwisizukulwana esitsha somxholo oveliswa ngokuzenzekelayo yitekhnoloji ye-AI emva kwe-UGC kunye ne-PGC.
Kungoko ke, akunzima ukufumanisa ukuba esinye sezizathu eziphambili zokuthandwa komzobo we-AI kukuba imodeli yomzobo we-AI inokuqonda ngokuthe ngqo oko kufakwa lulwimi ngumsebenzisi, kwaye idibanise ngokusondeleyo ukuqonda umxholo wolwimi kunye nokuqonda umxholo womfanekiso kwimodeli. I-ChatGPT ikwafumene ingqalelo njengemodeli yolwimi lwendalo olusebenzisanayo.
Akunakuphikwa ukuba, ngophuhliso olukhawulezayo lobukrelekrele bokwenziwa kwiminyaka yakutshanje, i-AIGC iqalisa utshintsho olutsha lweemeko zokusetyenziswa. Ividiyo yemizobo ye-AI, umzobo we-AI kunye neminye imisebenzi emeleyo yenza umfanekiso we-AIGC ubonakale kuyo yonke indawo kwividiyo emfutshane, kusasazo olubukhoma, ukusingatha kunye neqonga lepati, nto leyo eqinisekisa i-AIGC enamandla.
Ngokutsho kukaGartner, i-AI evelisayo iya kubalelwa kwi-10% yayo yonke idatha evelisiweyo ngo-2025. Ukongeza, uGuotai Junan ukwathe kwiminyaka emihlanu ezayo, i-10%-30% yomxholo womfanekiso isenokuveliswa yi-AI, kwaye ubungakanani bemarike obuhambelanayo bunokudlula i-60 yeebhiliyoni zeerandi.
Kuyabonakala ukuba i-AIGC ikhawulezisa ukuhlanganiswa okunzulu kunye nophuhliso kuzo zonke iinkalo zobomi, kwaye amathuba ophuhliso lwayo abanzi kakhulu. Nangona kunjalo, akunakuphikwa ukuba kusekho iingxabano ezininzi kwinkqubo yophuhliso lwe-AIGC. Uchungechunge lwezemizi-mveliso alugqibelelanga, iteknoloji ayivuthwanga ngokwaneleyo, imiba yobunini belungelo lokushicilela njalo njalo, ingakumbi malunga nengxaki "yokutshintsha i-AI endaweni yomntu", ukuya kuthi ga kwinqanaba elithile, uphuhliso lwe-AIGC luyathintelwa. Nangona kunjalo, uXiaobian ukholelwa ukuba i-AIGC ingangena kumbono woluntu, kwaye itshintshe iimeko zokusetyenziswa kwemizi-mveliso emininzi, kufuneka ibe neempawu zayo, kwaye amandla ayo ophuhliso kufuneka aphuhliswe ngakumbi.
Ixesha lokuthumela: Disemba-12-2022