I-ChatGPT iyisibonelo sokwenziwa kwengxoxo ye-intelligence
Kwaphuma izindaba ukuthi I-OpenAI ithuthukisa isofthiwe ekwazi ukubona ukuthi umbhalo wakho we-ChatGPT ukhiqizwe yini, lokhu ngemva nje kokuba izikhulu zemfundo zaseNew York City zimemezele ukuthi zizovimba ukufinyelela kwabafundi kuleli thuluzi ezikoleni zomphakathi.
Futhi wukuthi bakhiphe imibiko eyahlukene yokuthi yini abafundi abasebenzisa i-AI ukwenza umsebenzi wabo wesikole, benze othisha bacabange ukuthi iyithinta kanjani imfundo.
Abanye baye bazwakalisa ezabo ukukhathazeka ngendlela amamodeli olimi bangase bavumele abafundi ukuba bakhohlise.
Ngokunikezwa kwalokhu, i-OpenAI imemezele ukuthi isebenza ukuthuthukisa "ukunciphisa" okuzosiza abantu bathole umbhalo okhiqizwa ngokuzenzakalela yi-ChatGPT.
“Senze i-ChatGPT yatholakala njengokuhlola kuqala kocwaningo ukuze sifunde ekusetshenzisweni komhlaba wangempela, esikholelwa ukuthi kuyingxenye ebalulekile yokuthuthukisa nokukhipha amasistimu e-AI anamandla navikelekile. Sihlala sifaka impendulo nezifundo esizitholile,” kusho okhulumela inkampani.
Ngokusho kobufakazi boprofesa basenyuvesi, abafundi bathemba i-ChatGPT ukuxazulula umsebenzi wabo wesikole, ikakhulukazi ezifundweni.
“I-academy ayizange ibone ukuthi izofika. Ngakho kwasimangaza,” kusho uDarren Hudson Hick, umsizi kaprofesa wefilosofi eFurman University.
“Ngikubike kuFacebook, futhi abangani bami [othisha] babengathi, ‘yebo! Ngibambe eyodwa nami,'” engeza. Ngasekuqaleni kwale nyanga, kubikwa ukuthi u-Hick ucele ikilasi lakhe ukuthi libhale indatshana enamagama angama-500 ngesazi sefilosofi saseScotland sangekhulu le-XNUMX uDavid Hume kanye nendida eyesabekayo, ehlola ukuthi abantu bangathola kanjani injabulo entweni abayesabayo. , ukuze bayohlolwa ekhaya. Kodwa ngokusho kukaprofesa wefilosofi, enye yezincwadi eza kuye yayinezici ezithile "ezaziphawula" ukusetshenziswa kwe-AI ekuphenduleni komfundi "okuyisisekelo". U-Hick uyachaza ukuthi lokhu kungabonwa ngeso eliqeqeshiwe.
Ukwazi ukuhlukanisa ukubhala ngesandla ngesandla esikhiqizwe ngumuntu noma umshini izoshintsha indlela ezingasetshenziswa ngayo esikoleni. Izikole zingasebenzisa ngempumelelo ukuvinjelwa kokuhlolwa okwenziwe yi-AI,
Yebo, amamodeli olimi akhiqizayo angaba mahle, kodwa awazi ukuthi akhuluma ngani.
Okuhlaba umxhwele njengoba ukubhala okukhiqizwa yi-AI kuba yizihloko zezindaba ngezingqungquthela zasenyuvesi kanye nezikole ezivimbela amaphepha abhalwe ngomshini, nasi isikhumbuzo sokuthi abaqondi kahle ngokubhalwa komuntu kwangempela.
Futhi kungenxa yokuthi i-OpenAI ibilokhu ihlaba umxhwele i-inthanethi ngemizamo yayo yokuphindaphinda ubuhlakani bomuntu namakhono obuciko kusukela ngo-2015. Kodwa ngoNovemba odlule, inkampani yagcina ihambe kakhulu ngokwethulwa kwe-AI text generator ChatGPT. Abasebenzisi bethuluzi le-beta bathumele izibonelo zezimpendulo zombhalo okhiqizwe i-AI ekwazisweni obekubonakala kusemthethweni kangangokuthi bafaka ukwesaba ezinhliziyweni zothisha futhi benza i-Google yesaba ukuthi leli thuluzi lingabulala umsebenzi wabo wocwaningo.
Uma onjiniyela be-OpenAI bengakha i-bot engathayipha nayo noma engcono kunomuntu ojwayelekile, kunengqondo ukuthi bangakwazi futhi ukudala i-bot engcono kunomuntu ojwayelekile ekutholeni ukuthi umbhalo wakhiqizwa yi-AI.
Njengoba kushiwo okwamanje i-OpenAI isebenza ngesixazululo, okungenani amathuluzi amathathu okuthola asekhishiwe angasetshenziswa:
I-GPT-2 Exit Detector
Idemo eku-inthanethi yemodeli yesitshina sokuphuma ye-GPT-2 ikuvumela ukuthi unamathisele umbhalo ebhokisini futhi ubone ngokushesha amathuba okuthi umbhalo uthayiphe i-AI. Ngokusho kocwaningo lwe-OpenAI, leli thuluzi linezinga eliphezulu lokutholwa, kodwa "lidinga ukuhlanganiswa nezindlela ezisekelwe kumethadatha, ukwahlulela komuntu, kanye nemfundo yomphakathi ukuze liphumelele kakhulu."
I-GLTR
Ngenkathi i-OpenAI ikhipha i-GPT-2 ngo-2019, abantu baseMIT-IBM Watson AI Lab kanye ne-Harvard Natural Language Processing Group bahlangana ukuze bakhe i-algorithm ezama ukuthola ukuthi umbhalo uthayiphwe yini i-bot.
Umbhalo okhiqizwe ngekhompyutha ungase ubonakale ubhalwe ngumuntu, kodwa umlobi ongumuntu angase akhethe amagama angalindelekile. Ukusebenzisa indlela ethi "kwenza umuntu azi okukodwa", uma i-algorithm ye-GLTR ingabikezela igama elilandelayo emshweni, izothatha ngokuthi umusho ubhalwe i-bot.
GPZero
Phakathi nenkathi kaKhisimusi, u-Edward Tian wayematasa edala i-GPZero, uhlelo lokusebenza olungasiza ekunqumeni ukuthi umbhalo ubhalwe ngumuntu noma i-bot. Njengesifundiswa e-Princeton, u-Tian uyaqonda ukuthi osolwazi basenyuvesi bangaba kanjani nentshisekelo yokuthola ukukopela okusizwa yi-AI noma ukukopela okusizwa yi-AI.
U-Tian uthi ithuluzi lakhe likala ukungahleleki kwemisho ("buzz") kanye nokungahleliwe kukonke ("ukuqhuma") ukuze kubalwe amathuba okuthi umbhalo ubhalwe yi-ChatGPT. Kusukela ebhale kuTwitter nge-GPTZero ngoJan. 2, u-Tian uthi useke wathintwa ngonxiwankulu abafuna ukutshala imali, futhi uzothuthukisa izinguqulo ezibuyekeziwe maduze.
Umthombo: https://techcrunch.com