I-OpenAI, ilabhorethri ezinze e-San Francisco, California ethuthukisa ubuhlakani bokwenziwa obuhlanganisa izinhlobo ezinkulu zezilimi, umemezele amandla okudala izinguqulo zangokwezifiso ze-GPT-3, imodeli ekwazi ukukhiqiza ikhodi yohlobo lomuntu ngombhalo nenkulumo.
Ngayo kusukela manje onjiniyela bangasebenzisa ukushuna okuhle ukuze bakhe amamodeli e-GPT-3 yakhelwe okuqukethwe okuthile kwezinhlelo zakho zokusebenza namasevisi, okuholela emiphumeleni yekhwalithi ephezulu yayo yonke imisebenzi nemithwalo yomsebenzi, kuye ngenkampani.
Kulabo abangayijwayele i-GPT-3, kufanele wazi ukuthi lokhu iyimodeli yolimi ezenzakalelayo esebenzisa ukufunda okujulile ukukhiqiza imibhalo efana neyomuntu.
Yilokho imodeli yokubikezela ulimi yochungechunge lwesizukulwane sesithathu lwe-GPT-n oludalwe i-OpenAI, ilabhorethri yocwaningo lobuhlakani bokwenziwa esekelwe e-San Francisco eyakhiwe inkampani eyenza inzuzo i-OpenAI LP kanye nenkampani engumzali, inkampani engenzi nzuzo i-OpenAI Inc.
Kusuka kunoma yimuphi umlayezo wombhalo, njengomusho, i-GPT-3 ibuyisela umbhalo owengeziwe ngolimi lwemvelo.
Onjiniyela 'Bangahlela' i-GPT-3 ngokukubonisa izibonelo ezimbalwa noma 'ukwaziswa'.
"Siklame i-API ukuthi ibe lula kuwo wonke umuntu futhi ivumelane nezimo ngokwanele ukwenza amaqembu okufunda ngomshini akhiqize kakhudlwana," kusho i-OpenAI ngasekupheleni kukaMashi.
Ngalesi sikhathi, izicelo ezingaphezu kuka-300 zisebenzisa i-GPT-3 ezigabeni nasezimbonini ezahlukahlukene, kusukela ekukhiqizeni nasekufundiseni ngisho nokudala kanye nemidlalo.
La umthamo omusha wokucolisiswa kusilungiselelo se-GPT-3 sivumela amakhasimende ukuthi aqeqeshe i-GPT-3 ukubona iphethini ethile yomthwalo womsebenzi njengokukhiqizwa kokuqukethwe, ukuhlukaniswa kanye nokudidiyelwa kombhalo ngaphakathi kwemikhawulo yendawo ethile.
Umhlinzeki osebenzayo usebenzisa i-GPT-3 ukusiza izinkampani zisebenzise impendulo yamakhasimende. Isebenzisa idatha engahlelekile, isistimu ingakhiqiza imibiko efingqa impendulo yekhasimende nokusebenzisana. Ngokwenza ngokwezifiso i-GPT-3, i-Viable ibizokwazi ukukhulisa ukunemba kwemibiko yayo isuka ku-66% iye ku-90%.
Okufanayo kuya ku-Keeper Tax, ithuluzi elenza ukubalwa kwezimali zokuzisebenzela kube lula ngokuhlukanisa ngokuzenzakalelayo futhi kukhishwe idatha yomthwalo okhokhelwayo yemibiko yentela, ebhange noma ku-akhawunti yokukhokha. I-Keeper Tax isebenzisa i-GPT-3 ukuze ihumushe idatha yesitatimende sasebhange ukuze kutholwe izindleko okungenzeka zidonswe intela. Inkampani iyaqhubeka nokwenza ngcono i-GPT-3 ngedatha entsha isonto ngalinye ngokusekelwe ekusebenzeni kwangempela komkhiqizo wayo, igxile ezibonelweni lapho imodeli yehlele ngaphansi komkhawulo othile wokusebenza.
I-Los abathuthukisi bengeza cishe amasampula amasha angama-500 masonto onke ukucwenga imodeli. I-Keeper Tax ithi inqubo yokuhlanganisa ikhiqiza ukuthuthuka okungu-1% isonto nesonto.
«Into eyodwa esiqaphe kakhulu ngayo nesigcizelela kuyo ekuthuthukisweni kwethu le API ukwenza ukuthi ifinyeleleke kubathuthukisi abangenalo ngempela isizinda sokufunda ngomshini, ”kusho u-Rachel Lim, ilungu labasebenzi bezobuchwepheshe be-OpenAI. “Indlela ezibonakalisa ngayo ukuthi ungakwazi ukwenza ngokwezifiso ithempulethi ye-GPT-3 usebenzisa i-command line invocation. [Sithemba] ukuthi ngenxa yokufinyeleleka kwayo, singakwazi ukufinyelela isethi yabasebenzisi abahluke kakhulu abangaletha izinkinga zabo ezihluke kakhulu kwezobuchwepheshe. "
U-Lim uthi amandla okuthuthukisa e-GPT-3 nawo angaholela ekongiweni kwezindleko njengoba amakhasimende engalindela imvamisa ephezulu yemiphumela yekhwalithi engcono evela kumamodeli afakwe ngokunembile uma kuqhathaniswa nemodeli evamile ye-GPT-3. (Izindleko ze-OpenAI zokufinyelela i-API ngokusekelwe ku inombolo yamathokheni, noma amagama, akhiqizwa amamodeli.)
Nakuba i-OpenAI ine-premium kumamodeli acwengisisiwe, i-Lim ithi amamodeli amaningi ahlanjululwe adinga ukwaziswa okufushane okunamathokheni ambalwa, angaphinda asindise imali.
I-GPT-3 API ibilokhu itholakala esidlangalaleni kusukela ngo-2020. Unyaka ngaphambi kokwethulwa kwayo, abaklami bayo banqume ukungawenzi umphakathi umsebenzi wokuthuthukisa wenguqulo yangaphambilini, i-GPT-2, kucatshangelwa ukuthi lolu hlelo oluxakeke ngokufunda komshini lungase luphumelelise. kungaba yingozi uma iwela ezandleni zabantu abanonya.