I-ChatGPT ngumzekelo owenziwe ngobukrelekrele be-chatbot
Zakhutshwa iindaba ukuba I-OpenAI iphuhlisa isoftware enokubona ukuba imodeli yakho yeChatGPT yenziwe isicatshulwa, oku kamsinya nje emva kokuba amagosa ezemfundo kwisiXeko saseNew York ebhengeze ukuba aza kuvala ukufikelela kwabafundi kwesi sixhobo kwizikolo zikarhulumente.
Kwaye kukuba baye bakhupha iingxelo ezahlukeneyo zokuba yintoni abafundi abasebenzisa i-AI ukwenza umsebenzi wabo wasekhaya, baye benza ootitshala bacinga ngendlela eyichaphazela ngayo imfundo.
Abanye baye bavakalisa ezabo inkxalabo malunga nendlela iimodeli zolwimi banokuvumela abafundi ukuba bakope.
Ngenxa yoku, i-OpenAI ibhengeze ukuba isebenza ukuphuhlisa "ukunciphisa" okuza kunceda abantu babone isicatshulwa esiveliswa ngokuzenzekelayo yi-ChatGPT.
“Senze i-ChatGPT ifumaneke njengomboniso wophando ukuze sifunde kusetyenziso lwehlabathi lokwenyani, esikholelwa ukuba yinxalenye ebalulekileyo yokuphuhlisa nokuhambisa iinkqubo ze-AI ezikwaziyo nezikhuselekileyo. Sihlala sibandakanya ingxelo kunye nezifundo ezifundiweyo, ”utshilo isithethi senkampani.
Ngokobungqina obuvela koonjingalwazi baseyunivesithi, abafundi bathemba i-ChatGPT ukusombulula umsebenzi wabo wasekhaya, ingakumbi kuqheliselo.
“Iziko lemfundo alizange likubone liza. Ngoko yasothusa,” utsho uDarren Hudson Hick, usekela-njingalwazi wentanda-bulumko kwiYunivesithi yaseFurman.
"Ndayixela kuFacebook, kwaye abahlobo bam [abangutitshala] babenje, 'ewe! Ndibambe enye nam,'” wongeze watsho. Ekuqaleni kwale nyanga, u-Hick ucele iklasi yakhe ukuba ibhale isincoko samagama angama-500 kwisithandi sobulumko saseScotland senkulungwane ye-XNUMX uDavid Hume kunye ne-paradox eyoyikisayo, ephonononga ukuba abantu banokulufumana njani uyolo kwinto abayoyikayo. , kuvavanyo ekhaya. Kodwa ngokutsho kweNjingalwazi yefilosofi, enye yezincoko eza kuye yayineempawu ezithile "eziphawuleka" ukusetyenziswa kwe-AI kwimpendulo yomfundi "esisiseko". U-Hick uchaza ukuba oku kunokubonwa ngeso eliqeqeshiweyo.
Ukwazi ukwahlula ukubhala ngesandla ngesandla esiveliswe ngumntu okanye ngumatshini iya kutshintsha indlela ezinokusetyenziswa ngayo kwiziko lemfundo. Izikolo zinokukunyanzelisa ngokufanelekileyo ukuvalwa kovavanyo olwenziwe yi-AI,
Ewe, imifuziselo yolwimi oluvelisayo inokuba mihle, kodwa abayazi into abayithethayo.
Okuchukumisayo njengoko ukubhalwa kwe-AI kusenza izihloko eziphambili ngeenkomfa zaseyunivesithi kunye nezikolo ezivala amaphepha achwetheziweyo, nasi isikhumbuzo sokuba abaqondi kakuhle ukubhala komntu.
Kwaye kukuba i-OpenAI iye yachukumisa i-Intanethi ngeenzame zayo zokuphindaphinda ubukrelekrele bomntu kunye nobuchule bobugcisa ukusukela ngo-2015. Kodwa ngoNovemba ophelileyo, inkampani ekugqibeleni yahamba i-mega viral ngokusungulwa kwe-AI text generator ChatGPT. Abasebenzisi besixhobo se-beta bapapashe imizekelo yeempendulo ezibhaliweyo ezenziwe yi-AI kwimiyalezo ebonakala isemthethweni kangangokuba babethe uloyiko ezintliziyweni zootitshala bade benza uGoogle woyike ukuba isixhobo singabulala umsebenzi wabo wophando.
Ukuba iinjineli ze-OpenAI zinokwenza i-bot enokuthi ichwetheze okanye ibengcono kunomntu oqhelekileyo, kuyavakala ukuba banokwenza i-bot engcono kunomntu oqhelekileyo ekuboneni ukuba isicatshulwa senziwa yi-AI.
Ukusukela njengoko kukhankanyiwe okwangoku i-OpenAI isebenza kwisisombululo, ubuncinci izixhobo ezintathu zokufumanisa sele zikhutshiwe ezinokusetyenziswa:
I-GPT-2 Phuma kwi-Detector
Idemo ye-intanethi ye-GPT-2 imodeli ye-detector yokuphuma ikuvumela ukuba uncamathisele umbhalo kwibhokisi kwaye ngokukhawuleza ubone ukuba kunokwenzeka ukuba okubhaliweyo kuchwethezwe yi-AI. Ngokutsho kophando lwe-OpenAI, isixhobo sinezinga eliphezulu lokubona, kodwa "kufuneka zidityaniswe neendlela ezisekelwe kwi-metadata, isigwebo somntu, kunye nemfundo yoluntu ukuba isebenze kakhulu."
I-GLTR
Xa i-OpenAI ikhupha i-GPT-2 ngo-2019, abantu baseMIT-IBM Watson AI Lab kunye neqela leHarvard Natural Language Processing Group badibanela ukwenza i-algorithm ezama ukufumanisa ukuba umbhalo uchwethezwe yi-bot.
Umbhalo owenziwe ngekhompyutha usenokubonakala ngathi ubhalwe ngumntu, kodwa umbhali ongumntu usenokukhetha amagama angenakuqikelelwa. Ukusebenzisa indlela "ithatha umntu ukuba azi enye", ukuba i-algorithm ye-GLTR inokuqikelela igama elilandelayo kwisivakalisi, iya kucinga ukuba isivakalisi sibhalwe yi-bot.
GPZero
Ngexesha leKrisimesi, u-Edward Tian wayexakekile edala i-GPTZero, i-app enokunceda ukufumanisa ukuba umbhalo ubhalwe ngumntu okanye i-bot. Njengesifundiswa ePrinceton, uTian uyayiqonda indlela oonjingalwazi baseyunivesithi abanokuba nomdla ngayo ekubhaqeni ukukopa okuncediswa yi-AI okanye ukukopa okuncediswa yi-AI.
U-Tian uthi isixhobo sakhe silinganisa ukungakhethi kwezivakalisi ("buzz") kunye nokungakhethiyo ngokubanzi ("ukuqhuma") ukubala amathuba okuba okubhaliweyo kubhalwe yi-ChatGPT. Ukusukela kwi-tweeting malunga ne-GPTZero nge-2 kaJan, u-Tian uthi sele efikelwe ngoongxowankulu abafuna ukutyala imali, kwaye uya kuphuhlisa iinguqulelo ezihlaziyiweyo kungekudala.
Umthombo: https://techcrunch.com