Mhla ziyi-9 kuNcwaba, imodeli yokukhiqiza izithombe enekhodi ethi “mona-lisa-1” ivele ohlwini lokuhlolwa okuyimpumputhe lwe-AI Arena.
Umvumbululi u-AiBattle uthe ngemva kokuthumela izithombe ezikhiqizwe yile modeli ethuluzini elisemthethweni le-OpenAI lokuqinisekisa okuqukethwe, kutholwe izimpawu zemvelaphi ezihlobene ne-OpenAI. Ngakho-ke, le “mona-lisa-1” ithathwa njengemodeli yesizukulwane esisha ye-GPT-Image.
Kuze kube manje, i-OpenAI ayikakhiphi isimemezelo esisemthethweni, imibhalo yomkhiqizo noma incazelo yomphakathi ngale modeli, futhi ayikaqinisekisi ubudlelwano bayo obuqondile nemikhiqizo ekhona.
@TimJayas usebenzise imiyalelo efanayo ukuqhathanisa le modeli engaqondakali ne-GPT-Image 2. Imiphumela yokuhlolwa ikhombise ukuthi “mona-lisa-1” inemininingwane ecebile kwezinye izici.
Ngokwempendulo yabahloli abaningi, le modeli okwamanje ivela kuphela esigabeni sokuhlolwa okuyimpumputhe se-LM Arena, futhi ayikaxhunywa eziteshini zemikhiqizo esemthethweni efana ne-ChatGPT, i-OpenAI API noma i-Codex. Okunye ukuhlolwa kokuqhathanisa kukhombisa ukuthi “mona-lisa-1” inomehluko othile ekubukekeni kwemvelo, ekujuleni kwemininingwane nasekunciphiseni ukubukeka kokwenziwa okuvame ukubonakala ezithombeni ezikhiqizwe yi-AI, kodwa okunye ukuhlolwa kuphawule ukuthi eminye imiphumela esakhiqiziwe inomsindo noma izimpawu zokuphazamiseka.
Ngaphambi kokukhishwa ngokusemthethweni kwe-GPT-Image 2, nayo yake yavela enkundleni ye-Arena ngaphansi kwekhodi engaziwa ukuze ihlolwe ngokungaziwa. Okwamanje, ukwakheka kobuchwepheshe be-“mona-lisa-1”, isikhathi sokuqedwa kokuqeqeshwa kwayo kanye nohlelo lokukhishwa kwayo ngokusemthethweni kusazodalulwa yi-OpenAI.
Ikhasi elisemthethweni le-OpenAI Verify lichaza ngokusobala ukuthi leli thuluzi lingahlola ukuthi amafayela ezithombe noma omsindo anezimpawu zemvelaphi yokuqukethwe ezihlobene namathuluzi e-OpenAI yini, okuhlanganisa imethadatha ye-C2PA kanye ne-watermark ye-SynthID. Lihloselwe kakhulu okuqukethwe okukhiqizwe yi-ChatGPT, i-OpenAI API noma i-Codex.
I-OpenAI iphinde iveze ukuthi uma kutholwa izimpawu zemvelaphi ezisekelwayo, lokho kungakhombisa ukuthi okuqukethwe okukuleso simo kukhiqizwe noma kuthunyelwe ngaphandle kusetshenziswa ithuluzi le-OpenAI, kodwa imiphumela yokuqinisekisa yona ngokwayo ayikwazi ukuqinisekisa ukuthi iyiphi imodeli ethile esetshenzisiwe. Ngakho-ke, okwamanje akunakwenzeka ukuqinisekisa, ngokusekelwe kuphela ezimpawini zemvelaphi ezifana ne-SynthID, ukuthi “mona-lisa-1” iyimodeli eyalandela ngokuqondile i-GPT-Image 2.