- I-ChatGPT yenza amaphutha esakhiwo ekucabangeni, idatha, kanye nokuqonda okuvamile, naphezu kokubukeka kwayo "okuhlakaniphile".
- Amaphutha avela endleleni yokuqeqesha izibalo, ukucwasa kwedatha, kanye nokuntuleka kokuqonda kwangempela.
- Imisebenzi eminingi kanye nemisebenzi yezemfundo izoshintsha, kodwa i-AI izosebenza njengomsizi, hhayi njengomuntu ozongena esikhundleni sayo ngokuphelele.
- Ukusebenzisa i-ChatGPT ngokuqinisekiswa komthombo, izixwayiso ezinhle, kanye nokubuyekezwa komuntu kukuvumela ukuthi uyisebenzise ngaphandle kokuthatha izingozi eziningi.
Ubuhlakani bokwenziwa obukhiqizayo sebuyingxenye yempilo yethu yansuku zonke, kanti i-ChatGPT isibe ithuluzi elisetshenziswa kakhulu ekubhaleni, ekufingqeni, ekuhleleni izinhlelo, nasekucabangeni . Kodwa-ke, nakuba kungase kubonakale kumangalisa, akuphelele neze. Ukuthemba ngokungaqondile izimpendulo zayo kungaholela ezimangazweni ezingathandeki uma ungajwayelene nokulinganiselwa kwayo.
Yingakho ochwepheshe abaningi bephikelela ngokuthi ukuqonda imisebenzi lapho i-ChatGPT ingaphumeleli khona kubaluleke kakhulu njengokwazi ukuthi isetshenziswa kanjani . Akukhona ukudelela ubuchwepheshe, kodwa mayelana nokufunda ukuphila nokulinganiselwa kwabo: ukuthi buthembekile nini, ukuthi kufanele buthathwe nini izimpendulo zabo, nokuthi ungabusebenzisa kanjani ngaphandle kokulahlekelwa ukwahlulela kwabantu okugxekayo.
Kungani kubalulekile ukwazi imikhawulo ye-ChatGPT
Ngemuva kwesixhumi esibonakalayo sengxoxo kukhona uhlelo oluyinkimbinkimbi oluhlanganisa amamodeli amakhulu olimi nezindlela zangaphakathi ezibeka phambili isivinini noma ukucabanga kuye ngombuzo . Lesi "sinqumo" esizenzakalelayo asihlali sihambisana nezidingo zomsebenzisi: ngezinye izikhathi impendulo esheshayo ithatha indawo yokuqala noma inkinga idinga ukucabanga ngesinyathelo ngesinyathelo. Ungahlola ngokujulile ukuthi amapharamitha obuhlakani bokwenziwa athinta kanjani lokhu kuziphatha.
Ngaphezu kwalokho, ukuziphatha kwemodeli kuyashintsha ngokubuyekezwa ngakunye, okusho ukuthi i-ChatGPT ingaphendula ngendlela ehlukile kumbuzo ofanayo ngezikhathi ezahlukene . Okungeziwe kulokhu imikhawulo yomongo ezingxoxweni ezinde, lapho iqala khona "ukukhohlwa" izingxenye zomucu, kanye nezihlungi zokuphepha ezivimba izihloko ezibucayi noma ezisemthethweni ngokuphelele ngenxa yokuqapha okukhulu.
Lokhu kuhlanganiswa kwezici kusho ukuthi, empeleni, ulwazi lomsebenzisi lungashintsha kusuka kokugqamile kuya kokukhungathekisayo ngemizuzwana embalwa . Umehluko phakathi komphumela owusizo nowokubhubhisa uvame ukuba semephini umsebenzisi angakwazi yini ukubona ukuthi i-AI iyaphambuka nini nokuthi angakulungisa kanjani.
Izifundo eziningana zakamuva kanye nezinhlolovo zibonisa ukuthi ingxenye enkulu yabasebenzisi ayizethembi ngokugcwele izimpendulo ze-ChatGPT . Kuhlolo lwe-OCU, abantu abaningi bakhulume "ngokuntuleka kokwethemba" ulwazi olutholiwe njengesizathu esiyinhloko sokungalisebenzisi nsuku zonke, naphezu kokulizama.
Ngenxa yalokho, ochwepheshe bezobunhloli bokwenziwa kanye nezemfundo batusa ukusebenzisa i-ChatGPT njengomthombo owengeziwe futhi ungalokothi ube ngumlamuli weqiniso . Ukuqinisekisa idatha neminye imithombo, okungenani ukuqonda okuyisisekelo ngendaba, nokwamukela ukuthi ulwazi lungakhiwa (“imibono engaqondakali” edumile, kulezi zinsuku, kubalulekile.
Amaphutha avamile kanye nemikhawulo yobuchwepheshe ye-ChatGPT
Izincwadi zobuchwepheshe zibhale ukuthi lezi zinhlelo zivame ukukhiqiza izimpendulo ezinengqondo kodwa ezingamanga , ikakhulukazi uma umbuzo ungacacile, ucacile kakhulu, noma uphambuka kudatha abayibonile ngesikhathi sokuqeqeshwa. Inkinga ukuthi amagama ayakholisa kangangokuthi kunzima ukubona iphutha uma umsebenzisi engajwayelene nesihloko. Amathuluzi afana nezandiso zokuthola okuqukethwe okukhiqizwe yi-AI angasiza ekuboneni ezinye zalezi zimpendulo ezingamanga.
Ochwepheshe abanjengoSolwazi uJosep Curto bagcizelela ukuthi amanye amaphutha avame kakhulu ajulile futhi athinta ukuthembeka okuphelele kohlelo . Lokhu kufaka phakathi izincazelo eziyiphutha zamaqiniso aqinisekisiwe, izimpendulo ezingaphelele, izincazelo eziqanjiwe "zenqubo yokucabanga" yomodeli, kanye nokusebenza okunganembi kangako kwezinye izilimi ngaphandle kwesiNgisi.
Ngaphezu kwalokho, yize kuqaliswa ukuthuthukiswa, imikhawulo yokuziphatha kanye neyokuphepha ayilinganiswa kahle ngaso sonke isikhathi . Ngezinye izikhathi ivimba izicelo ezisemthethweni ngenkathi ivumela eziyinkinga, noma iveza izimpendulo ezigcwele ubandlululo oluzuzwe kudatha yokuqeqeshwa, kungaba ngokwesiko, ngokwezepolitiki, noma ngokobulili.
Naphezu kwakho konke, ucwaningo oluqhubekayo ekufundeni okujulile kanye nokucubungula ulimi lwemvelo luholele ezinguqulweni zakamuva, njenge-GPT-4, ekufinyeleleni ukusebenza okucishe kube ngokomuntu emisebenzini eyinkimbinkimbi kwizibalo, izinhlelo, umthetho, ezokwelapha kanye ne-psychology. Kodwa ukusebenza okuphezulu akusho ukungaphuthelwa, futhi kusalokhu kunezici ezingaboni.
Imisebenzi engu-7 lapho i-ChatGPT yehluleka khona (noma ingathembekile)
Nakuba i-ChatGPT ingaba usizo kakhulu ezimweni eziningi, kunezinhlobo ezithile zemisebenzi lapho amathuba okuthola impendulo engalungile, echemile, noma edukisayo anda kakhulu . Kunconywa ukuthi ujwayelane nalezi zimo ukuze uqaphele kakhulu.
1. Ukucabanga okunengqondo kanye nezibalo eziyinkimbinkimbi
Uma inkinga idinga izibalo eziboshwe ngeketanga, imiboniso enengqondo, noma ubufakazi obusemthethweni bezibalo , i-ChatGPT ivame ukungaphumeleli. Ingenza amaphutha emisebenzini elula, yeqe izinyathelo ezibalulekile, noma iveze imiphumela engalungile njengelungile.
Lokhu kubonakala kakhulu ezivivinyweni ezinezinyathelo eziningi eziphakathi, njenge-combinatorics, probability, advanced algebra, noma i-geometry . Ngisho noma umbhalo ubonakala uphikiswana kahle, amaphutha amancane esinyathelweni esisodwa angenza umphumela wonke ungasebenzi, futhi imodeli ayinandlela "yokuhlola" ngempela ukucabanga kwayo.
2. Ukuqonda okuvamile, ukuqonda, kanye nomongo womuntu
I-AI ihumusha lokho etshelwa khona ngokoqobo, ngakho-ke ihluleka ukuqonda ukuhlekisa, izincazelo ezimbili, ukubhuqa, noma izinkomba zamasiko eziqondile kakhulu . Lokhu kuphumela ezimpendulweni, kumuntu, ngokusobala "ezingekho endaweni yazo."
Uma umsebenzi udinga uzwela lwangempela, ukuqonda okujulile ngokomzwelo, noma ulwazi olusebenzayo ngomhlaba wangempela , imodeli ilingisa nje amaphethini olimi, kodwa ngaphandle kokuhlangenwe nakho komuntu siqu. Kungase kuzwakale sengathi kunozwela, kodwa akuzizwa noma yini noma akuboni izimo njengoba umuntu ebengazibona. Ukuze uthuthukise ithoni futhi ulungise izimpendulo, kuyalulekwa ukwenza ngezifiso i-ChatGPT futhi ulungise izitayela nezindima.
3. Inkumbulo yesikhathi eside kanye nokuqhubeka kwephrojekthi
Nakuba kungase kubonakale sengathi "kuyakhumbula" ingxoxo, i-ChatGPT inomongo olinganiselwe futhi ingalahlekelwa izingxenye zomlando ngaphandle kwesixwayiso . Izingxoxo ezinde kakhulu noma amaphrojekthi athatha isikhathi eside avame ukuba nokuphelelwa yinkumbulo.
Lokhu kuthinta ngqo imisebenzi efana nokubhala izincwadi, imibiko emide, noma amaphrojekthi okuhlela amaseshini amaningi . Uma okuqukethwe kungagcinwa futhi kungaphathwa ngaphandle, kunengozi yokulahlekelwa ukuvumelana, ukuphinda izinqumo esezenziwe, noma ukuziphikisa ngaphandle kokuba imodeli ikwazi lokho.
4. Kucatshangwa ukuthi ukonga isikhathi okukugcina kukubiza kakhulu
Abantu abaningi baphendukela ku-ChatGPT ukuze bathole isivinini, kodwa uma izimpendulo ziqukethe amaphutha amancane noma zingenalo ulwazi olubalulekile, ukuzibuyekeza nokuzilungisa kungathatha isikhathi eside kunokuzenza kusukela ekuqaleni . Lokhu kubonakala kakhulu emisebenzini yobungcweti edinga izinga eliphezulu lekhwalithi.
Ezindaweni ezifana nemibiko yobuchwepheshe, imibhalo yezomthetho, okuqukethwe kwezemfundo, noma ikhodi yokukhiqiza , isikhathi esengeziwe esichithwa ekuqinisekiseni singayiqeda ngokuphelele inzuzo ecatshangelwayo yokusebenzisa i-AI uma kungekho nqubo yokubuyekeza eqinile.
5. Idatha ephelelwe yisikhathi kanye nokuntuleka kokuxhumana ngesikhathi sangempela
Amamodeli afana ne-GPT-3 noma i-GPT-4 aqeqeshwa kusetshenziswa isikhathi esithile sedatha, ngakho-ke awanakho ukufinyelela okuqondile kuwebhu ukuze ahlole ulwazi lwesikhathi sangempela . Kuye ngenguqulo, ulwazi lwawo lungase lukhawulelwe onyakeni othile.
Lokhu kusho ukuthi, maqondana nemicimbi yamanje, izinguquko zomthetho zakamuva, intuthuko yesayensi, noma izindaba ezisanda kuvela , izimpendulo zingase zibe ziphelelwe yisikhathi noma zingamanga. Uma umsebenzisi engahloli imithombo ebuyekeziwe, iphutha alibonakali.
6. Ukusebenza okuguquguqukayo kanye nokugcwala
Ngesikhathi sokufunwa okuphezulu, ikakhulukazi kuma-akhawunti wamahhala, ikhwalithi kanye nesivinini sezimpendulo kungathinteka . Imodeli inganikeza izimpendulo ezingenamininingwane eminingi, inqamule izimpendulo ngaphambi kwesikhathi, noma inikeze izixazululo ezingasho lutho.
Lokhu kuhlukahluka kusho ukuthi akusiyona njalo indawo ethembekile yemisebenzi ebalulekile enezinsuku zokugcina eziqinile . Ukuthembela ngokuphelele kusevisi ngezikhathi ezimatasa ngaphandle kohlelo lokusekela kungaba yingozi yokusebenza kwamabhizinisi nochwepheshe.
7. Imikhawulo, izihlungi kanye nokubandlulula
Ngenxa yezizathu zokuphepha, i-ChatGPT ifaka izihlungi zokuvimbela okuqukethwe okulimazayo, okungekho emthethweni, noma okubucayi kakhulu . Inkinga ukuthi lezi zinhlelo zokuhlola aziphelele futhi ngezinye izikhathi zivimba izicelo ezisemthethweni noma ziphendule ngokungacacile ngenxa yokuqaphela ngokweqile.
Noma kunjalo, abacwaningi abaningana bakhombisile ukuthi amamodeli asengakhiqiza izimpendulo ngobandlululo lobuhlanga, ubulili, noma imibono , okubonisa ubandlululo olukhona kudatha yokuqeqeshwa. Amacala abhalwe phansi ama-bot aqeqeshwe ngamasu afanayo agcina ekhiqiza okuqukethwe okubandlululayo. Ukuhlolwa kokuziphatha kwama-chatbot kusiza ukuqonda nokunciphisa lokhu kubandlulula.
Amaphutha ayisisekelo: kungani i-ChatGPT yenza amaphutha amaningi kangaka
Ekuhlaziyeni la maphutha, izazi zefilosofi kanye nochwepheshe be-AI bakhomba imikhawulo yesakhiwo: I-ChatGPT ayicabangi, ayiqondi, noma ayinalo ulwazi ngalokho ekushoyo . Ubuhlakani bayo obubonakalayo buvela ekuxhumaneni okukhulu kwezibalo, hhayi ekucabangeni kwangempela.
Ngokwesibonelo, uProfesa uNed Block ukhombe ekuhlolweni kwamamodeli ezithombe ezikhiqizayo lapho abasebenzisi becelwa ukudweba iwashi elibonisa izikhathi ezinjengo-12:03 noma u-6:28, futhi cishe njalo iwashi livela ngo-10:10 . Leso sikhathi sibusa ezithombeni zokukhangisa eziku-inthanethi ngoba siyathandeka futhi asisithezi ilogo.
Lokhu kuziphatha kubonisa indlela amamodeli avame ukuphinda ngayo amaphethini avame kakhulu kudatha yawo yokuqeqesha, noma ngabe aphikisana nemiyalelo ethile . "Abaqondi" umqondo wesikhathi; bamane baphinda ukucushwa okuvame kakhulu ngokwezibalo.
Kwenzeka okufanayo uma ucelwa isithombe somuntu ebhala ngesandla sakhe sobunxele . Amamodeli amaningi agcina ekhiqiza abantu abasebenzisa isandla sokudla, ngoba izibonelo eziningi kudatha nazo zisebenzisa isandla sokudla. Ukuthola umuntu osebenzisa isandla sokudla omelelwe kahle kudinga ukusetshenziswa okuqhubekayo kwemiyalelo ethile, futhi noma kunjalo akusebenzi ngaso sonke isikhathi.
Lolu hlobo lwamaphutha esakhiwo lusho ukuthi ngisho nabathuthukisi abakwazi ukulungisa ngokuphelele ukucwasa okujulile . Ukuqeqeshwa kokuqinisa ngempendulo yabantu kuyasiza, kodwa ukuqinisa ngesandla zonke izimo ezingaba khona (isikhathi ngasinye sewashi, yonke indawo yokubhala, njll.) akunakwenzeka.
Ezingeni lombhalo, inkinga ihunyushwa ibe izincazelo eziqiniseka kakhulu zezinto imodeli "engazi " ngempela. Bangase bachaze, isinyathelo ngesinyathelo, indlela okucatshangwa ukuthi abakaze bayisebenzise, noma bakhe izinkomba zemfundo kanye nezixhumanisi nomoya okholisayo ngokuphelele.
Indlela i-ChatGPT eqeqeshe ngayo nokuthi kungani lokho kuthonya amaphutha akhe
Amamodeli afana ne-GPT-3 kanye ne-GPT-4 asekelwe kusakhiwo se-Transformer, futhi aqeqeshwe ngombhalo omningi ovela kwi-inthanethi : amakhasi ewebhu, izincwadi, izihloko zesayensi, izindaba kanye nezinye izinsiza zomphakathi ezifana ne-Wikipedia noma i-Common Crawl.
Le nqubo inezigaba ezimbili eziyinhloko. Okokuqala, ukuqeqeshwa okukhulu kwangaphambi kokufunda lapho imodeli ifunda khona ukubikezela igama elilandelayo ngezigidi zemisho . Kulesi sigaba, ayifundiswa “uhlelo lolimi” olucacile, kodwa kunalokho ithola amaphethini ngokugcwalisa izikhala embhalweni—uhlobo lokuzivocavoca okukhulu kokuqedela amagama.
Ngemuva kwalokho, ukulungiswa kahle kwenzelwa imisebenzi ethile, njengokuhumusha , ukufingqa, ukuphendula imibuzo, noma ingxoxo yengxoxo . Lokhu kuhilela ukusebenzisa amaphuzu edatha amancane, aqondile kakhulu, anelebula ukufundisa imodeli ukuthi hlobo luni lwempendulo olubhekwa njengolufanele kumongo ngamunye.
Endabeni ye-ChatGPT, kwasetshenziswa futhi ukufunda kokuqinisa impendulo yomuntu (i-HFRL) . Abaqeqeshi babantu bahlole izimpendulo eziningana ezingaba khona zemodeli kokufakwayo okufanayo futhi bazibeka ohlwini kusukela kokuhle kakhulu kuya kokubi kakhulu. Ngalolu lwazi, uhlelo lufunde ukukhetha imiphumela enesilinganiso esiphezulu.
Le ndlela ithuthukisa ikhwalithi yezimpendulo ezibonwayo, kodwa ayisusi inkinga eyisisekelo: imodeli isantula ukuqonda okunencazelo kanye nokufinyelela okuqondile eqinisweni . Ithuthukisa kuphela amathuba ayo okuzwakala iwusizo kakhudlwana, inenhlonipho, futhi iqiniseka, okwenza ngezinye izikhathi imibono engekho ibe yingozi nakakhulu.
Ngaphezu kwalokho, yize ukuhlanzwa kwedatha, ukuhlelwa kabusha, kanye namasu okuhlola abantu kusetshenziswa ukunciphisa ukucwasa kanye nemibono engavamile ekhona emibhalweni yokuqeqesha , akunakwenzeka ukuyihlunga ngokuphelele. Ngakho-ke, izinkampani zincoma ukuqapha okuqhubekayo komkhiqizo, ikakhulukazi ezisetshenzisweni ezibucayi.
Umthelela we-ChatGPT ezifundweni nasemsebenzini
Kusukela yethulwa, i-ChatGPT ivuselele impikiswano mayelana nokuthi i-AI ekhiqizayo izothinta kanjani imisebenzi yabantu kanye nohlelo lwemfundo . Izinkampani ezinkulu zobuchwepheshe ziyiveza njenge-"co-pilot co-pilot" noma umsizi, kodwa izifundo eziningi zikhomba izinguquko ezinkulu emisebenzini eminingi.
Ucwaningo oluvela emanyuvesi afana neNew York University kanye ne-University of Pennsylvania, kanye nokuhlaziywa okuvela ku-OpenAI kanye nezinye izinhlangano, lubonisa ukuthi imisebenzi esekelwe ezilimini (ukubhala, ukuhumusha, ukuhlaziya umbhalo, imibhalo, ukubalwa kwezimali okuyisisekelo) isengozini enkulu yokwenza izinto ngokuzenzakalela okungaphelele.
Emakethe yemisebenzi, abadayisi bezingcingo, oprofesa bezilimi nezincwadi zaseyunivesithi, othisha bomlando nomthetho, abahumushi, abasebenzi bokuphatha, kanye nezindima ezithile zokubhala baye babonakala njengabasengozini kakhulu . Lokhu akusho ukuthi bonke bazonyamalala, kodwa kusho ukuthi imisebenzi yabo eminingi ingasekelwa kakhulu yi-AI.
Ezinye izifundo zithole imiphumela emihle ecacile. Ekuhlolweni kwabathuthukisi besofthiwe, isibonelo, labo abasebenzisa abasizi bokubhala amakhodi abasekelwe kumamodeli baqede imisebenzi ngokushesha kufika ku-55% kunalabo abasebenza ngaphandle kosizo. Ukukhiqiza emisebenzini evamile kuyanda, okuvumela isikhathi esiningi sokuzinikela ezinkingeni eziyinkimbinkimbi.
Kwezemfundo, impikiswano ishubile kakhulu. Izifundo zibonisa ukuthi kunzima kakhulu ukuhlukanisa ukuthi indatshana emfushane yabhalwa ngumfundi noma i-ChatGPT , ngisho nakothisha abanolwazi nababhali abangochwepheshe. Kwezinye izivivinyo, othisha nochwepheshe behlulekile ukuthola umbhali weqiniso.
Ngenxa yalokho, othisha abaningi sebeqala ukuphakamisa ukuhlolwa okwengeziwe ngomlomo, umsebenzi wasekilasini, kanye nemisebenzi ebonisa inqubo yokucabanga yomfundi , hhayi nje umkhiqizo wokugcina. Abanye basikisela ukusebenzisa i-ChatGPT njengethuluzi lokufundisa: indawo yokuqala umfundi okumele ayigxeke, ayilungise, ayiqhathanise nemithombo yangempela, futhi ayithuthukise.
Izindlela ezinhle kakhulu zokusebenzisa i-ChatGPT ngaphandle kokuwela ezingibeni zayo
Naphezu kwazo zonke lezi zinkinga, i-ChatGPT iwusizo kakhulu uma inikwe indima efanele. Isihluthulelo ukuhlanganisa ukwakheka okuhle okusheshayo kanye nokuqapha okuqinile komuntu umphumela wayo , ikakhulukazi uma kunemiphumela yezemfundo, yezomthetho, noma yobungcweti.
Isincomo sokuqala esiyisisekelo ukuqinisekisa njalo idatha, izibalo, izinkomba, namagama afanele ngaphambi kokuzisebenzisa. Akwanele ukuthi into "izwakale kahle": kufanele uhlole imithombo ethembekile (izihloko zesayensi, imithetho esemthethweni, izincwadi zokubhekisela, izizindalwazi ezikhethekile, njll.).
Kunconywa futhi ukuthi ungasekeli izinqumo ezibucayi empendulweni eyodwa evela kumodeli . Uma indaba ibalulekile, kuwukuhlakanipha ukufuna imibono eminingi (kwamanye amathuluzi, ochwepheshe babantu, noma imibhalo) bese usebenzisa ingxoxo kuphela njengethuluzi lokusekela noma lokwenza ama-draft.
Ukwakheka kwemiyalelo kunomthelela omkhulu: izixwayiso ezicacile, eziqondile, nezihlelwe kahle zinciphisa ukungaqondakali futhi zivame ukukhiqiza imiphumela ewusizo kakhulu . Ukucela umsebenzisi ukuthi achaze ukucabanga kwakhe isinyathelo ngesinyathelo, afake izibonelo eziphikisayo, noma aveze imikhawulo engaba khona kungasiza ekutholeni amaphutha.
Okokugcina, ukusebenzisa i-ChatGPT njengomkhiqizi wamathempulethi, izinhlaka, ama-draft, kanye nohlu lwemibono umuntu azosebenza kuyo kunamandla kakhulu , esikhundleni sokwamukela umbhalo wokugcina. Lokhu kusindisa isikhathi engxenyeni yomshini, kodwa okuqukethwe kokugcina kusalungiswa futhi kuqinisekiswa ngumuntu.
Okuhlangenwe nakho kweminyaka yamuva kubonisa ukuthi ubuhlakani bokwenziwa obukhiqizayo busebenza kahle kakhulu uma buphathwa njengomsizi onamandla kodwa onamaphutha . Ngeso elibukhali, ukwahlulela okuhle ekukhetheni imisebenzi, kanye nohlelo lokubuyekeza oluqinile, buba umngane obalulekile kunokuba usongo oluthule.
