Ukubuyela kundoqo, impumelelo ye-AIGC kubunye yindibaniselwano yezinto ezintathu:

1. I-GPT yi-replica ye-neuron yomntu

I-GPT AI emelwe yi-NLP yi-algorithm ye-neural network yekhompyutha, eyona nto iphambili yayo ikukulinganisa uthungelwano lwe-neural kwi-cortex ye-cerebral cortex.

Ukusetyenzwa kunye nokucinga okukrelekrele kolwimi, umculo, imifanekiso, kunye nolwazi lokungcamla yonke imisebenzi eqokelelwe ngumntu.

ingqondo njenge "protein yekhompyuter" ngexesha lokuzivelela kwexesha elide.

Ke ngoko, i-GPT ngokwemvelo yeyona ndlela ifanelekileyo yokulinganisa ukusetyenzwa kolwazi olufanayo, oko kukuthi, ulwimi olungacwangciswanga, umculo, kunye nemifanekiso.

Indlela yokusetyenzwa kwayo ayikokuqonda intsingiselo, kodwa yinkqubo yokusulungekiswa, ukuchonga nokudibanisa.Oku kubi kakhulu

into edidayo.

Ii-algorithms zentetho yangaphambili zesemantic zokuqaphela imodeli yegrama kunye nedathabheyisi yentetho, emva koko yenza imephu intetho kwisigama,

emva koko wabeka isigama kwisiseko sedatha yegrama ukuqonda intsingiselo yesigama, kwaye ekugqibeleni wafumana iziphumo zokuqaphela.

Ukuqondwa kakuhle kolu lwazi “lwengqiqo” lusekwe kwi-syntax lujikeleza malunga ne-70%, efana ne-ViaVoice.

I-algorithm eyaziswa yi-IBM kwi-1990s.

I-AIGC ayikho malunga nokudlala ngolu hlobo.Undoqo wayo ayikokukhathalela igrama, kodwa kunokuba kusekwe i-algorithm yenethiwekhi ye-neural evumela ukuba

ikhomputha ukubala udibaniso olunokwenzeka phakathi kwamagama ahlukeneyo, aludibaniso lwe-neural, hayi udibaniso lwesemantic.

Ngokufana nokufunda ulwimi lwethu lweenkobe sisebancinane, ngokwemvelo salufunda, kunokuba sifunde “umxholo, isivisa, into, isenzi, umphelelisi,”

kunye nokuqonda umhlathi.

Lo ngumzekelo wokucinga we-AI, ukuqaphela, kungekhona ukuqonda.

Oku kukwabaluleke kakhulu kwi-AI kuzo zonke iimodeli zendlela yakudala - iikhompyuter akufuneki ziwuqonde lo mbandela kwinqanaba elinengqiqo,

kodwa endaweni yoko chonga kwaye uqaphele unxulumano phakathi kolwazi lwangaphakathi, kwaye ke uyazi.

Umzekelo, imo yokuqukuqela kwamandla kunye noqikelelo lweegridi zamandla zisekwe kufaniso lothungelwano lwamandla eklasiki, apho imodeli yemathematika

Indlela yokusebenza isekiwe kwaye emva koko idityaniswe kusetyenziswa ialgorithm ye-matrix.Kwixesha elizayo, kusenokungabi yimfuneko.I-AI iya kuchonga ngokuthe ngqo kwaye iqikelele a

ipateni ethile yemodyuli esekwe kubume benodi nganye.

Okukhona kukho iindawo ezininzi, kokukhona ithandwa kancinci i-algorithm ye-matrix yakudala, kuba ubunzima be-algorithm buyanda ngenani

iindawo kunye nokuqhubela phambili kwejometri kwanda.Nangona kunjalo, i-AI ikhetha ukuba ne-concurrency ye-node enkulu kakhulu, kuba i-AI ilungile ekuchongeni kunye

ukuqikelela ezona ndlela zothungelwano ezinokwenzeka.

Nokuba luqikelelo olulandelayo lweHamba (iAlphaGO inokuqikelela amashumi alandelayo amanqanaba, ngokunokubalwa okunokwenzeka kwinyathelo ngalinye) okanye uqikelelo lwemodal.

yeenkqubo zemozulu ezintsonkothileyo, ukuchaneka kwe-AI kuphezulu kakhulu kuneemodeli zoomatshini.

Isizathu sokuba kutheni igridi yamandla ngoku ayifuni i-AI kukuba inani leendawo ezikwi-220 kV nangaphezulu kothungelwano lwamandla olulawulwa liphondo.

Ukuthunyelwa akukukhulu, kwaye iimeko ezininzi zisetelwe ukulungelelanisa kunye nokunciphisa i-matrix, ukunciphisa kakhulu ubunzima bokubala

imodeli yomatshini.

Nangona kunjalo, kwinqanaba lokuhamba kwamandla womnatha wokuhambisa amandla, ejongene namashumi amawaka okanye amakhulu amawaka eendawo zamandla, iindawo zomthwalo, kunye nemveli.

I-algorithms ye-matrix kuthungelwano olukhulu losasazo alunamandla.

Ndiyakholelwa ukuba ukuqatshelwa kwepateni ye-AI kwinqanaba lomnatha wokusabalalisa kuya kwenzeka kwixesha elizayo.

2. Ukuqokelela, uqeqesho, kunye nokuveliswa kolwazi olungacwangciswanga

Isizathu sesibini sokuba kutheni i-AIGC yenze impumelelo kukuqokelelwa kolwazi.Ukusuka kwi-A/D yokuguqulwa kwentetho (i-microphone+PCM

isampuli) ukuya kwi-A/D yokuguqulwa kwemifanekiso (i-CMOS + imephu yendawo yombala), abantu baqokelele idatha ye-holographic kwimbonakalo kunye nokuphicotha.

amasimi ngeendlela ezinexabiso eliphantsi kakhulu kule minyaka imbalwa idlulileyo.

Ngokukodwa, ukuthandwa okukhulu kweekhamera kunye nee-smartphones, ukuqokelelwa kwedatha engacwangciswanga kwintsimi ye-audiovisual yabantu.

malunga neendleko eziphantse zero, kwaye ukuqokelela okuqhumayo kolwazi lombhalo kwi-Intanethi ngundoqo kuqeqesho lwe-AIGC - iiseti zedatha yoqeqesho azibizi.

Lo mzobo ungentla ubonisa ukukhula kwedatha yehlabathi, ebonisa ngokucacileyo intsingiselo ye-exponential.

Oku kukhula okungahambelaniyo kokuqokelelwa kwedatha kusisiseko sokukhula okungahambelaniyo kwezakhono ze-AIGC.

KODWA, uninzi lwezi datha ziyidatha ye-audio-visual data engalungiswanga, eqokelelwe ngexabiso elinguziro.

Kwintsimi yamandla ombane, oku akunakufezekiswa.Okokuqala, uninzi lweshishini lamandla ombane lucwangciswe kwaye luqingqelwe kancinci idatha, efana

I-voltage kunye ne-current, eziziiseti zedatha yeengongoma zechungechunge lwexesha kunye ne-semi structured.

Iiseti zedatha yezakhiwo kufuneka ziqondwe ziikhompyuter kwaye zifuna "ulungelelwaniso", olufana nolungelelwaniso lwesixhobo - amandla ombane, angoku, kunye nedatha yamandla.

yokutshintsha kufuneka ilungelelaniswe kule nodi.

Okukhathaza ngakumbi kukulungelelaniswa kwexesha, efuna ulungelelwaniso lombane, langoku, kunye namandla asebenzayo kunye asebenzayo ngokusekwe kwisikali sexesha, ukuze

ukuchongwa okulandelayo kunokwenziwa.Kukwakho nezalathiso eziya phambili nasemva, ezilulungelelwaniso lwesithuba kwiiquadrants ezine.

Ngokungafaniyo nedatha yokubhaliweyo, engadingi ulungelelwaniso, umhlathi uphoswa nje kwikhompyuter, echonga unxulumano lolwazi olunokwenzeka.

ngokwayo.

Ukuze ulungelelanise lo mbandela, njengokulungelelaniswa kwezixhobo zedatha yokusasazwa kweshishini, ulungelelwaniso lufuneka rhoqo, kuba okuphakathi kunye

Inethiwekhi yokusasazwa kwamandla ombane aphantsi iyongeza, icime, kwaye ilungisa izixhobo kunye nemigca yonke imihla, kwaye iinkampani zegridi zichitha iindleko ezinkulu zomsebenzi.

Njengo "uchazo lwedatha," iikhompyuter azikwazi ukwenza oku.

Okwesibini, iindleko zokufumana idatha kwicandelo lamandla liphezulu, kwaye iinzwa zifunwa endaweni yokuba nefowuni ephathwayo ukuthetha nokuthatha iifoto.”

Ngalo lonke ixesha i-voltage iyancipha ngenqanaba elinye (okanye ubudlelwane bokusabalalisa amandla buyancipha kwinqanaba elinye), i-investment sensor investment iyanda

ubuncinane ngomyalelo omnye wobukhulu.Ukufikelela kwicala lomthwalo (isiphelo se-capillary), lutyalo-mali olukhulu lwedijithali.

Ukuba kuyimfuneko ukuchonga imo edlulayo yegridi yamandla, i-high-precision high-frequency samples iyadingeka, kwaye ixabiso liphezulu.

Ngenxa yexabiso eliphezulu ngokugqithisileyo lokufunyanwa kwedatha kunye nokulungelelaniswa kwedatha, igridi yamandla okwangoku ayikwazi ukuqokelela ngokwaneleyo engeyiyo emgceni.

ukukhula kolwazi lwedatha ukuqeqesha i-algorithm ukufikelela kubunye be-AI.

Ungakhankanyi ukuvuleka kwedatha, akunakwenzeka ukuba ukuqaliswa kwe-AI yamandla ukufumana ezi datha.

Ngoko ke, phambi kwe-AI, kuyimfuneko ukuxazulula ingxaki yeesethi zedatha, ngaphandle koko ikhowudi ye-AI jikelele ayikwazi ukuqeqeshwa ukuvelisa i-AI enhle.

3. Ukuphumelela ngamandla okubala

Ukongeza kwii-algorithms kunye nedatha, impumelelo yobunye be-AIGC ikwayimpumelelo kumandla okubala.Ii-CPU zemveli azikho

ilungele i-neuronal computing enkulu ngaxeshanye.Kukusetyenziswa ngokuchanekileyo kweGPUs kwimidlalo ye-3D kunye neemuvi ezenza ukuhambelana okukhulu.

indawo edadayo+yokusasaza ikhompuyutha enokwenzeka.Umthetho kaMoore wehlisa ngakumbi iindleko zokubala ngeyunithi nganye yamandla okubala.

Igridi yamandla AI, umkhwa ongenakuthintelwa kwixesha elizayo

Ngokudityaniswa kwenani elikhulu le-photovoltaic esasazwayo kunye neenkqubo zokugcina amandla, kunye neemfuno zesicelo

indawo yomthwalo wezityalo zamandla ezinenyani, kunyanzelekile ukuba kuqhutywe uqikelelo lomthombo kunye nomthwalo weenkqubo zothungelwano losasazo loluntu kunye nomsebenzisi

ukuhanjiswa (micro) iinkqubo zegridi, kunye nexesha lokwenyani lokuhamba kwamandla okusasaza (micro) iinkqubo zegridi.

Ubunzima bokubala kwicala lothungelwano losasazo ngenene luphezulu kunolo lweshedyuli yothumelo lothungelwano.Nokuba kurhwebo

entsonkothileyo, kunokubakho amashumi amawaka ezixhobo zomthwalo kunye namakhulu okutshintsha, kunye nemfuno ye-AI esekwe kwigridi encinci / yokusebenza kwenethiwekhi yosasazo.

ulawulo luya kuvela.

Ngexabiso eliphantsi labenzi boluvo kunye nokusetyenziswa ngokubanzi kwezixhobo zombane zamandla ezifana ne-slid-state transformers, i-slid-state switch, kunye ne-inverters (abaguquli),

ukudityaniswa kwenzwa, i-computing, kunye nolawulo kumda wegridi yamandla nayo ibe yinto entsha.

Ngoko ke, i-AIGC yegridi yamandla yikamva.Nangona kunjalo, into efunekayo namhlanje akuyikukhupha ngokukhawuleza i-algorithm ye-AI ukwenza imali,

Endaweni yoko, qala ujongane nemiba yolwakhiwo lweziseko zedatha efunwa yi-AI

Ekuphakameni kwe-AIGC, kufuneka kubekho ukucinga okuzolileyo okwaneleyo malunga nenqanaba lesicelo kunye nekamva lamandla e-AI.

Okwangoku, ukubaluleka kwamandla e-AI akubalulekanga: umzekelo, i-algorithm ye-photovoltaic enokuchaneka kokuchaneka kwe-90% ibekwe kwimarike yendawo.

kunye nomda wokuphambuka kwi-5%, kwaye ukuphambuka kwe-algorithm kuya kucima yonke inzuzo yokurhweba.

Idatha ngamanzi, kwaye amandla okubalwa kwe-algorithm yitshaneli.Njengoko kusenzeka, kuya kuba njalo.