Kwenzekeni
Intatheli yeNew Scientist uMatthew Sparkes ithi i-ChatGPT ikhiqize isixazululo enkingeni yamadayisi yabadlali abayisithupha ngemuva kwemizuzu eyi-13 yokucabanga. Umcwaningi u-Eric Harshbarger uhlole izinombolo wazithola zilungile. Ukwakhiwa kusebenzisa amadayisi amahlanu anezinhlangothi ezingu-720 kanye nefa elilodwa elinezinhlangothi ezingu-20, okwenza kusebenze ngokwezibalo kodwa kungenzeki.
I-New Scientist ibika ukuthi u-Sparkes ucele inguqulo yamahhala ye-ChatGPT ukuthola amadayisi ayisithupha kubadlali abayisithupha ukuze bonke abashonile babe nethuba elilinganayo lokuwina futhi akukho ukuphonswa okungaholela ekulinganeni. Isicelo silandela umzamo weminyaka eyishumi wongoti bezibalo ukwakha isethi enezigaba ezinhlanu ezifanayo. U-Sparkes uthi i-chatbot ikhiqize impendulo ngemuva kwemizuzu ecishe ibe ngu-13, ilandela iziyalezo ezimbili ezicacisayo. U-Harshbarger, umcwaningi weNyuvesi yase-Auburn ohileleke kulo msebenzi wokufa kwabantu abahlanu, wahlola ukwakhiwa kwezinombolo futhi watshela i-New Scientist ukuthi yayilungile.
Ukwakhiwa okubikiwe akuwona umkhiqizo owaziwa kakhulu noma osebenzayo wokudlala. Amadayisi amahlanu anezinhlangothi ezingama-720 kanti elesithupha linezinhlangothi ezingama-20. I-New Scientist ithi i-Harshbarger kanye nabahlanganyeli base bevele baguqule umphumela wabo wamadayisi amahlanu bawenza ukwakhiwa kwamadayisi ayisithupha lapho ukufa ngakunye kunezinhlangothi ezingama-360. Iwebhusayithi yabacwaningi futhi iqukethe isixazululo samadayisi amahlanu kanye nendlela yokwelula izixazululo kumadayisi amaningi. I-New Scientist ithi ukusesha akutholanga mkhondo wempendulo eqondile yamadayisi ayisithupha, kodwa ababambiqhaza abakwazanga ukunquma ukuthi i-chatbot izitholele lezo zingxenye zomphakathi futhi yawuqeda umsebenzi noma yakhiqiza isisombululo ngokuzimela.
Imininingwane yomthombo: newscientist.com ↗
Kungani kubalulekile
Umbiko unikeza isibonelo esiphathekayo, esihlolwe ngokuzimele sesistimu ye-AI yenhloso evamile ehlanganisa ulwazi lwezibalo lube yisixazululo esivumelekile. Ayiqinisekisi ukuthi imodeli idale umqondo omusha wezibalo, noma ukuthi ingaxazulula izinkinga ezinzima ngokuthembekile ngaphandle kokuqondisa komuntu. Okushiwo okungokoqobo ukuthi ama-chatbot angasiza abacwaningi ukuthi bakhiqize izinto zokwakha, kuyilapho abantu besadinga ukuziqinisekisa futhi bahlulele ukuthi ziwusizo yini.
Lokhu ukuboniswa kwekhono eliwusizo ngoba i-chatbot ibonakala ixhume ukwakhiwa kwezibalo ngendlela eyaziwayo yokungenisa futhi yanikeza isandiso esivumelekile ngesiNgisi esilula. Ukuhlolwa kwezinombolo kuka-Harshbarger kuwubufakazi obuzwakalayo bokuthi impendulo bekungekhona okuphumayo okungahlangani, kodwa akufani nobufakazi obusemthethweni noma ukuhlola okulawulwayo kuzo zonke izinkinga eziningi.
Umbiko uphinde ubonise umngcele womphumela. Impendulo ayilungile futhi ayisebenzi ngokomzimba, futhi umthombo awuhlinzeki ngokuhlolwa okuzimele kwenqubo yokucabanga ye-chatbot. I-New Scientist icaphuna u-OpenAI ka-Sébastien Bubeck ebikezela inqubekelaphambili eqhubekayo, kodwa lokho umbono wochwepheshe kunobufakazi balokhu kuhlolwa. I-athikili ithi amasistimu amanje angathola izixazululo noma izibonelo eziphikisana nezinkinga ezikhona kodwa awazange abonise ukudalwa kwemiqondo emisha yezibalo noma izinkambu.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
Which component of an AI application is the machine-learning model itself?
Ongakubuka ngokulandelayo
Imibuzo eyinhloko engaxazululiwe ukuthi ingabe umphumela ungaphinda ukhiqizwe ngokuzimele, noma ngabe ubufakazi obuphelele nokulotshiwe kuyatholakala, nokuthi ingakanani impendulo evela ezintweni esezishicilelwe kakade ku-inthanethi. I-athikili ichaza i-chatbot njengemahhala kodwa ayibonisi imodeli yangempela, inguqulo, imikhawulo yokusetshenziswa, noma ukuthi amandla ayatholakala yini kabanzi.
Ukuhlola okuqinile kungadinga umbhalo ophelele wokwaziswa nempendulo, ukuqinisekiswa okusemthethweni kokwakhiwa, nokuhlolwa kochwepheshe bezibalo ababengayazi impendulo elindelekile. Kungasiza futhi ukuthola ukuthi imodeli isebenzise iwebhusayithi yabacwaningi, ukuthi kungakanani ukwaziswa obekudingeka, nokuthi umphumela ungaphinda ukhiqizwe ngokufinyelela okufanayo kwamahhala.
Umbuzo obanzi ukuthi ingabe i-AI ingakwazi yini ukujwayela kusuka kuzithako zezibalo ezaziwayo iye ezinkingeni ezintsha ngempela, kunokuxazulula okuhlukile okunezindlela zayo ezifanele sezivele ziku-inthanethi. Kuze kube lokho kuboniswa, umbiko usekela ukusetshenziswa okucophelelayo kwama-chatbot njengabasizi bokukhiqiza nokuhlola imibono yamakhandidethi, hhayi njengokumiselela ubufakazi, ukwahlulela, noma ukubhalwa kwezibalo.