Sasisele sithethile ngesinye isihlandlo UDALL-E. Ngesi sihlandlo kuvela inguqulelo yayo yesithathu. I-DALL-E3 ligama lenguqulelo entsha yobukrelekrele bokwenziwa I-OpenAI eyenza imifanekiso esuka kwisicatshulwa. Olu lutshintsho lwe-DALL-E, oluthiwe thaca ngoJanuwari ka-2021 kwaye sele lothuse umhlaba ngokukwazi ukuvelisa imifanekiso yeengcamango ezahlukeneyo. njengepenguin enomnqwazi okanye iavokhado emile okwesitulo. I-DALL-E 3 iphucula ngokubonakalayo ukusebenza kunye nobuchule be-predecessor yayo, inika imifanekiso engokoqobo, eneenkcukacha kunye nehambelanayo kunye nombhalo onikiweyo.
Kwakhona, ngokwendalo idibanisa neChatGPT, i-GPT-3-based chatbot evumela ukuba uxoxe kunye nobukrelekrele bokwenziwa kwaye uyicele ukuba yenze imifanekiso ngokwemiyalelo yethu. Kweli nqaku siza kukuxelela indlela iDALL-E 3 isebenza ngayo, zeziphi izinto ezintsha ezizisayo malunga ne-DALL-E, luhlobo luni lwemifanekiso enokuthi luyenze kwaye yintoni echaphazela le teknoloji kwixesha elizayo loyilo kunye nokunxibelelana.
Isebenza njani iDALL-E 3?
I-DALL-E3 Yimodeli yobukrelekrele eyenziweyo esekelwe kuyo iinethiwekhi ezenziweyo ze-neural, ngokukodwa kwizinto ezibizwa ngokuba yi-transformers, ezikwazi ukucubungula ukulandelelana kwedatha, njengesicatshulwa okanye imifanekiso, kunye nokufunda ubudlelwane phakathi kwabo.
Lo mzekelo uqeqeshwe ngenani elikhulu leeperi ze-text-image, ekhutshwe kwi-intanethi, ukufunda ukudibanisa iikhonsepthi ezibonakalayo namagama. Ngale ndlela, xa enikwe isicatshulwa, uyakwazi ukuvelisa umfanekiso owubonisayo, esebenzisa ubuchule bakhe kunye nokucinga.
Fumana zombini isicatshulwa kunye nomfanekiso njengomjelo wedatha enye, yenziwe ngamathokheni aphezulu angama-1280. Umqondiso yiyo nayiphi na isimboli yesigama esicacileyo; Umzekelo, unobumba ngamnye wealfabhethi luphawu. Isigama sika-DALL-E 3 ineempawu zokubhaliweyo kunye nomfanekiso. Umbhalo umelwe kusetyenziswa ubuninzi beempawu ze-256 ezifakwe ngekhowudi nge-BPE (i-Byte Pair Encoding), kwaye umfanekiso umelwe usebenzisa ii-tokens ze-1024 ezifakwe ngekhowudi. I-VQ-VAE (iVector Quantized Variational Autoencoder).
I-DALL-E 3 iqeqeshelwa ukusebenzisa indlela ephezulu yokunokwenzeka, ebandakanya ukuvelisa zonke iithokheni, enye emva komnye, ukwandisa amathuba okuba ngamnye anikwe ngaphambili. Ngale ndlela, iDALL-E 3 ingenza umfanekiso ukusuka ekuqaleni, okanye uvelise ngokutsha nayiphi na inxalenye yomfanekiso osele udlulela kwikona esezantsi ekunene, ukuba nje ungqinelana nokubhaliweyo.
Iza neziphi iindaba?
I-DALL-E 3 ithatha ukuqhubela phambili okukhulu xa kuthelekiswa neDALL-E kwimiba emininzi. Okokuqala, i-DALL-E 3 inesisombululo esiphezulu kunye nomgangatho kwimifanekiso eyenzayo. Ngelixa i-DALL-E idale imifanekiso ye Iiphikseli ezingama-256 × 256, I-DALL-XO 3 yenza imifanekiso ye Iiphikseli ezingama-512 × 512, ekuvumela ukuba uqonde ngakumbi iinkcukacha kunye nokuthungwa.
Okwesibini, iDALL-E 3 ine-a ukuqonda okukhulu kunye nokuchaneka xa utolika isicatshulwa esinikiweyo. Iyakwazi ukubamba ngcono i-nuances kunye neenkcukacha zombhalo, kunye nobudlelwane phakathi kwezinto ezenza umfanekiso. Umzekelo, unokwenza imifanekiso enombhalo ngaphakathi, njengeepowusta okanye iilebhile, ngokuhlonipha ulwimi kunye nefomathi yesicatshulwa. Unokwenza imifanekiso enamalungu omzimba womntu anokwenyani kwaye alinganayo, njengezandla okanye iinyawo.
Okwesithathu, DALL-XO 3 inokuhlanganiswa okukhulu kunye nokulula yokusetyenziswa enkosi kuqhagamshelo lwayo neChatGPT. I-ChatGPT yi-OpenAI's chatbot esekwe kwi-GPT-3, eyona modeli yolwimi iphambili kwihlabathi, ekuvumela ukuba uncokole nobukrelekrele bokwenziwa kwaye uyicele ukuba yenze izinto. Ngokudibanisa ne-ChatGPT, i-DALL-E 3 ungafumana imiyalelo eneenkcukacha ezininzi kunye nemifanekiso ecacileyo yokudala imifanekiso, kunye nokubonelela ngempendulo yendalo kunye neyolwelo kumsebenzisi.
Loluphi uhlobo lwemifanekiso enokwenziwa nguDALL-E 3?
I-DALL-E3 inokudala imifanekiso yeengqikelelo ezininzi ezahlukeneyo ezinokuchazwa ngolwimi lwendalo. Eminye imizekelo yile:
- Imifanekiso yezinto ze-anthropomorphized okanye izilwanyana, oko kukuthi, eneempawu zobuntu. Ngokomzekelo, ikati kwi-suit kunye ne-tie, okanye indlovu kwiiglasi kunye nomnqwazi.
- Imifanekiso yezinto ezixutyiweyo okanye izilwanyana, oko kukuthi, kunye neempawu ezidibeneyo zeentlobo ezimbini okanye ngaphezulu. Umzekelo, inja enamaphiko ebhabhathane, okanye inyoka enentloko yengonyama.
- Imifanekiso yezinto ezilungisiweyo okanye izilwanyana, oko kukuthi, ngeempawu ezitshintshiweyo okanye ezongezelelweyo. Umzekelo, imoto enamavili etshizi, okanye intyatyambo enamagqabi eglasi.
- Imifanekiso yezinto ezicingelwayo okanye izilwanyana, oko kukuthi, azikho ngokwenene. Ngokomzekelo, i-unicorn epinki, okanye inyoka yomlilo.
- Imifanekiso yemiboniso eyintsomi okanye imbonakalo-mhlaba, oko kukuthi, azihambelani nayo nayiphi na indawo yokwenene. Umzekelo, isixeko esidadayo esibhakabhakeni, okanye ihlathi elinemichilo.
- Imifanekiso esuka kuguqulo okanye ukuguqulwa kwemifanekiso ekhoyo, oko kukuthi, batshintsha inkalo ethile yomfanekiso wokuqala. Umzekelo, ukutshintsha umbala weenwele zomntu okanye amehlo, okanye ukongeza okanye ukususa into emfanekisweni.
I-DALL-XO 3 ineziphi iimpembelelo?
I-DALL-XO 3 ngumzekelo wesakhono esikhulu ubukrelekrele bokwenziwa obunabo kwinkalo uyilo kunye nonxibelelwano. Nge-DALL-E 3, ithuba lokudala imifanekiso eyenzelwe umntu kunye neyokuqala ivula ngokubhala nje ibinzana, elithi inokuba nezicelo ezininzi ezisebenzayo kunye nobuchule.
Umzekelo, i-DALL-E 3 inokusetyenziselwa uku:
- Yenza imifanekiso yeencwadi, iimagazini okanye iiblogi.
- Yenza iilogo okanye iipowusta kwiimpawu okanye iziganeko.
- Yenza i-avatar okanye i-emojis kwiintanethi zentlalo okanye imidlalo.
- Yenza i-memes okanye izitikha ukwabelana nabahlobo.
- Yenza imizobo okanye iiprototypes kwiiprojekthi zobugcisa okanye zobungcali.
- Yenza imifanekiso efundisayo okanye ulwazi lokuchaza iikhonsepthi ezintsonkothileyo.
Nangona kunjalo, i-DALL-E 3 ikwabeka imingeni ethile kunye nemingcipheko ekufuneka ithathelwe ingqalelo. Kwelinye icala, iDALL-E 3 inokuchaphazela umsebenzi kunye nokuqatshelwa kwabayili abangabantu kunye namagcisa, abanokubona ubuchule babo kunye nemvelaphi yabo isongelwa ngumatshini. Ngakolunye uhlangothi, i-DALL-E 3 inokuququzelela ukudala kunye nokusasazwa komxholo ongeyonyani okanye olahlekisayo, njengama-deepfakes okanye iindaba ezingeyonyani, ezinokuba neziphumo ezibi kuluntu.
Intelekelelo yakho, ngoku ngaphandle kwemiqobo
I-DALL-E 3 yinguqulelo entsha ye I-OpenAI yobukrelekrele bokwenziwa eyenza imifanekiso esuka kwisicatshulwa. I-DALL-E 3 iphucula umgangatho kunye nokuchaneka kwemifanekiso eyenzayo, kunye nokuhlanganiswa kwayo ne-ChatGPT. Unokwenza imifanekiso engakholelekiyo yeengqikelelo ezininzi ezahlukeneyo ezinokuchazwa ngolwimi lwendalo. I-DALL-XO 3 ine amandla amakhulu oyilo kunye nonxibelelwano, kodwa ikwabangela imingeni ethile kunye nemingcipheko ekufuneka ithathelwe ingqalelo.