Kuyini ubuhlakani bokwenziwa okukhiqizayo: busebenza kanjani, izinzuzo kanye nezingozi

I-Generative AI yisihloko sengxoxo yezobuchwepheshe esishisa kakhulu sango-2023.

Iyini i-generative intelligence yokufakelwa, isebenza kanjani futhi imayelana nani? Ake sikubone ndawonye kulesi sihloko

Iyini intelligence yokwenziwa?

I-Generative AI iwuhlobo lobuchwepheshe bobuhlakani bokwenziwa obuchaza kabanzi izinhlelo zokufunda zomshini ezingakhiqiza umbhalo, izithombe, ikhodi, noma ezinye izinhlobo zokuqukethwe.

Amamodeli we ubuhlakani bokwenziwa kokukhiqiza ziya ngokuya zifakwa kumathuluzi e-inthanethi futhi chatbot okuvumela abasebenzisi ukuthi bathayiphe imibuzo noma imiyalelo endaweni yokufaka, lapho imodeli ye-AI izokhiqiza impendulo efana nomuntu.

I-generative intelligence yokwenziwa isebenza kanjani?

Amamodeli we ubuhlakani bokwenziwa kokukhiqiza basebenzisa inqubo yekhompyutha eyinkimbinkimbi eyaziwa ngokuthi deep learning ukuhlaziya amaphethini avamile nokuhlelwa kumasethi amakhulu wedatha bese usebenzisa lolu lwazi ukuze udale imiphumela emisha nephoqelelayo. Amamodeli akwenza lokhu ngokuhlanganisa amasu okufunda omshini aziwa ngokuthi amanethiwekhi e-neural, agqugquzelwa ngokukhululekile indlela ubuchopho bomuntu obucubungula ngayo futhi buhumushe ulwazi bese lufunda kukho ngokuhamba kwesikhathi.

Ukwenza isibonelo, ukondla imodeli ye ubuhlakani bokwenziwa kokukhiqiza Ngokulandisa okukhulu, ngokuhamba kwesikhathi imodeli izokwazi ukubona nokukhiqiza kabusha izici zendaba, njengesakhiwo sesakhiwo, abalingiswa, izindikimba, amathuluzi okuxoxa nokunye.

Amamodeli we ubuhlakani bokwenziwa kokukhiqiza ziba yinkimbinkimbi kakhulu njengoba idatha abayitholayo futhi ikhiqiza ukwanda, futhi ngenxa yamasu we deep learning futhi of inethiwekhi ye-neural ngezansi. Ngenxa yalokho, kulapho isifanekiso esikhiqiza khona okuqukethwe okwengeziwe ubuhlakani bokwenziwa kokukhiqiza, lapho imiphumela yako ikholisa futhi ifane neyomuntu.

Izibonelo ze-AI yokukhiqiza

Ukuduma kweubuhlakani bokwenziwa kokukhiqiza yaqhuma ngo-2023, ikakhulukazi ngenxa yezinhlelo I-ChatGPT e I-DALL-E di I-OpenAI. Ngaphezu kwalokho, ukuthuthuka okusheshayo kobuchwepheshe ubuhlakani bokufakelwa, njengokucutshungulwa kolimi lwemvelo, wenze iubuhlakani bokwenziwa kokukhiqiza kufinyeleleka kubathengi nabadali bokuqukethwe ngesilinganiso.

Izinkampani ezinkulu zobuchwepheshe zisheshe zagxumela ku-bandwagon, i-Google, i-Microsoft, i-Amazon, i-Meta nezinye zonke zifakela amathuluzi azo okuthuthukisa. ubuhlakani bokwenziwa kokukhiqiza ezinyangeni ezimbalwa.

Kunamathuluzi amaningi ubuhlakani bokwenziwa kokukhiqiza, nakuba umbhalo kanye namamodeli okukhiqiza izithombe cishe aziwa kakhulu. Amamodeli we ubuhlakani bokwenziwa kokukhiqiza ngokuvamile bathembele kumsebenzisi ohlinzeka ngomlayezo obaqondisa ekukhiqizeni okukhiphayo okufunayo, kungaba umbhalo, isithombe, ividiyo noma ucezu lomculo, nakuba lokhu kungenjalo ngaso sonke isikhathi.

Izibonelo zamamodeli obuhlakani bokwenziwa okukhiqizayo

• I-ChatGPT: imodeli yolimi lwe-AI eyakhiwe yi-OpenAI engaphendula imibuzo futhi ikhiqize izimpendulo ezifana nezomuntu ezivela emiyalweni yombhalo.
• KUSUKA-E 3: enye imodeli ye-AI evela ku-OpenAI engakha izithombe nomsebenzi wobuciko kusuka emiyalweni yombhalo.
• I-Google Bard: I-chatbot ye-Google ekhiqizayo ye-AI kanye nembangi ye-ChatGPT. Iqeqeshelwe imodeli yolimi olukhulu lwe-PaLM futhi ingaphendula imibuzo futhi ikhiqize umbhalo kusuka ekwazisweni.
• UClaudius 2 : I-San Francisco-based Anthropic, eyasungulwa ngo-2021 ngabacwaningi bangaphambili be-OpenAI, imemezele inguqulo yakamuva yemodeli yayo ye-Claude AI ngoNovemba.
• Uhambo lwaphakathi : Ithuthukiswe ilebhu yocwaningo esekwe eSan Francisco iMidjourney Inc., le modeli ye-AI ihumusha imiyalelo yombhalo ukuze kukhiqizwe izithombe nomsebenzi wobuciko, ofana ne-DALL-E 2.
• IGitHub Copilot : ithuluzi lekhodi elinamandla e-AI eliphakamisa ukuqedwa kwekhodi endaweni yokuthuthukiswa kwe-Visual Studio, i-Neovim, ne-JetBrains.
• Isiqephu 2: Imodeli yolimi enkulu yomthombo ovulekile we-Meta ingasetshenziswa ukudala amamodeli e-AI engxoxo yama-chatbots nabasizi ababonakalayo, abafana ne-GPT-4.
• I-xAI: Ngemuva koxhaso lwe-OpenAI, u-Elon Musk ushiye iphrojekthi ngoJulayi 2023 futhi wamemezela lolu hlelo olusha lwe-AI lokukhiqiza. Imodeli yayo yokuqala, i-Grok engahloniphi, yaphuma ngoNovemba.

Izinhlobo zamamodeli e-AI akhiqizayo

Kunezinhlobo ezahlukahlukene zamamodeli e-AI akhiqizayo, ngayinye yakhelwe izinselele ezithile nemisebenzi. Lezi zingahlukaniswa ngobubanzi zibe izinhlobo ezilandelayo.

Transformer-based models

Amamodeli asuselwa ku-Transformer aqeqeshwa kumasethi amakhulu edatha ukuze aqonde ubudlelwano phakathi kolwazi olulandelanayo, njengamagama nemisho. Isekelwe ngu deep learning, lawa mamodeli e-AI athambekele ekubeni azi kahle i-NLP futhi aqonde ukwakheka nesimo solimi, okuwenza afanelekele kahle imisebenzi yokukhiqiza umbhalo. I-ChatGPT-3 ne-Google Bard ziyizibonelo zamamodeli e-AI akhiqizayo asuselwa ku-transformer.

Generative adversarial networks

Ama-GAN akhiwe amanethiwekhi amabili e-neural aziwa ngokuthi ijeneretha nokucwasa, empeleni asebenzisana ukuze adale idatha ebukeka iyiqiniso. Njengoba igama liphakamisa, indima yejeneretha iwukukhiqiza okukhiphayo okukholisayo okufana nesithombe esisuselwe esiphakamisweni, kuyilapho umbandlululi esebenza ukuze ahlole ubuqiniso besithombe esishiwo. Ngokuhamba kwesikhathi, ingxenye ngayinye ithuthuka ezindimeni zayo, ithola imiphumela ekholisayo. Kokubili i-DALL-E ne-Midjourney ziyizibonelo zamamodeli e-AI akhiqizayo asuselwa ku-GAN.

Variational autoencoders

Ama-VAE asebenzisa amanethiwekhi amabili ukuze atolike futhi akhiqize idatha: kulokhu iyisishumeki kanye nesikhiphi khodi. Isifaki khodi sithatha idatha yokufaka futhi siyiminyanise ibe ifomethi eyenziwe lula. Idekhoda ibe isithatha lolu lwazi olucindezelweyo bese ilwakha kabusha lube into entsha efana nedatha yasekuqaleni, kodwa akufani neze.

Isibonelo kungaba ukufundisa uhlelo lwekhompuyutha ukwenza ubuso bomuntu kusetshenziswa izithombe njengedatha yokuqeqeshwa. Ngokuhamba kwesikhathi, lolu hlelo lufunda ukwenza izithombe zobuso babantu zibe lula ngokuzinciphisa zibe izici ezimbalwa ezibalulekile, njengobukhulu nokuma kwamehlo, ikhala, umlomo, izindlebe, njll., bese uzisebenzisa ukwakha ubuso obusha.

Multimodal models

Amamodeli e-Multimodal angaqonda futhi acubungule izinhlobo eziningi zedatha ngesikhathi esisodwa, njengombhalo, izithombe, nomsindo, okubavumela ukuthi bakhe imiphumela eyinkimbinkimbi kakhulu. Isibonelo kungaba imodeli ye-AI engakhiqiza isithombe ngokusekelwe ekwazisweni kombhalo, kanye nencazelo yombhalo yokwaziswa kwesithombe. KUSUKA-E 2 e I-GPT-4 ngu-OpenAI ziyizibonelo zamamodeli we-multimodal.

Izinzuzo zobuhlakani bokwenziwa obukhiqizayo

Emabhizinisini, ukusebenza kahle kuyinzuzo ephoqelela kakhulu ye-AI ekhiqizayo ngoba ingenza amabhizinisi akwazi ukwenza imisebenzi ethile futhi agxilise isikhathi, amandla kanye nezinsiza ezinhlosweni zamasu ezibaluleke kakhulu. Lokhu kungaholela ezindlekweni eziphansi zabasebenzi, ukwanda kokusebenza kahle kanye nemininingwane emisha yokuthi izinqubo ezithile zebhizinisi ziyasebenza noma cha.

Kochwepheshe nabadali bokuqukethwe, amathuluzi e-AI akhiqizayo angasiza ekukhiqizeni imibono, ukuhlela nokuhlelwa kokuqukethwe, ukuthuthukiswa kwenjini yokusesha, ukumaketha, ukubandakanya izethameli, ucwaningo nokuhlela, nokunye okungaba nokunye. Futhi, inzuzo eyinhloko ehlongozwayo ukusebenza kahle ngoba amathuluzi e-AI akhiqizayo angasiza abasebenzisi behlise isikhathi abasichitha emisebenzini ethile ukuze batshale amandla abo kwenye indawo. Sekushiwo lokho, ukugadwa ngesandla nokulawulwa kwamamodeli e-AI akhiqizayo kuhlala kubaluleke kakhulu.

Amakesi okusebenzisa i-Generative AI

I-Generative AI isithole isisekelo emikhakheni eminingi yezimboni futhi ikhula ngokushesha ibe izimakethe zezentengiselwano nezabathengi. UMcKinsey uyalinganisela ukuthi, ngo-2030, imisebenzi okwamanje ethatha cishe u-30% wamahora okusebenza e-United States ingaba i-othomathikhi, ngenxa yokusheshisa kobuhlakani bokwenziwa obukhiqizayo.

Ensizakalweni yamakhasimende, ama-chatbots anamandla e-AI kanye nabasizi ababonakalayo basiza izinkampani ukuthi zinciphise izikhathi zokuphendula futhi zibambe ngokushesha imibuzo evamile yamakhasimende, zehlise umthwalo kubasebenzi. Ekuthuthukisweni kwesofthiwe, amathuluzi e-AI akhiqizayo asiza onjiniyela ikhodi ngokuhlanzeka nangempumelelo ngokubuyekeza ikhodi, ukugqamisa iziphazamisi, nokuphakamisa izixazululo ezingaba khona ngaphambi kokuba zibe izinkinga ezinkulu. Ngaleso sikhathi, ababhali bangasebenzisa amathuluzi e-AI akhiqizayo ukuze bahlele, babhale futhi babuyekeze izindatshana, izindatshana, nomunye umsebenzi obhaliwe, nakuba ngokuvamile kuba nemiphumela exubile.

Imikhakha yezicelo

Ukusetshenziswa kwe-AI ekhiqizayo kuyehluka kusuka embonini kuya embonini futhi kusungulwe kakhulu kwezinye kunezinye. Amacala okusetshenziswa kwamanje nokuhlongozwayo ahlanganisa okulandelayo:

• Impilo: i-generative AI ibhekwa njengethuluzi lokusheshisa ukutholakala kwezidakamizwa, kuyilapho amathuluzi afana nalokhu I-AWS HealthScribe bavumela odokotela ukuthi babhale ukubonisana nesiguli futhi balayishe ulwazi olubalulekile kumarekhodi abo ezokwelapha ngogesi.
• Ukumaketha kwedijithali: abakhangisi, abakhangisi namaqembu ezentengiselwano angasebenzisa i-AI ekhiqizayo ukuze bakhe imikhankaso yomuntu siqu futhi balungise okuqukethwe okuvumelana nezintandokazi zabathengi, ikakhulukazi uma kuhlanganiswa nedatha yokuphatha ubudlelwano bekhasimende.
• Umyalelo: Amanye amathuluzi okufundisa aseqala ukuhlanganisa i-AI ekhiqizayo ukuze kuthuthukiswe izinto zokufunda eziqondene nawe ezibhekelela izitayela zokufunda zomfundi ngamunye.
• Ezezimali: I-Generative AI ingelinye lamathuluzi amaningi ngaphakathi kwezinhlelo zezezimali eziyinkimbinkimbi ukuhlaziya amaphethini ezimakethe kanye nokulindela ukuthambekela kwezimakethe zamasheya, futhi isetshenziswa kanye nezinye izindlela zokubikezela ukusiza abahlaziyi bezezimali.
• Imvelo: kwisayensi yezemvelo, abacwaningi basebenzisa amamodeli obuhlakani bokwenziwa abakhiqizayo ukuze babikezele amaphethini wesimo sezulu futhi balingise imiphumela yokushintsha kwesimo sezulu.

Izingozi nemikhawulo yobuhlakani bokwenziwa okukhiqizayo

Okukhathaza kakhulu mayelana nokusetshenziswa kwamathuluzi e-AI akhiqizayo - futhi ikakhulukazi lawo afinyeleleka emphakathini - amandla awo okusabalalisa ulwazi olungamanga nokuqukethwe okuyingozi. Umthelela walokhu ungaba obanzi futhi ube nzima, kusukela ekuqhubekiseni phambili inkolelo-mbono, inkulumo enenzondo kanye nemibono elimazayo kuye ekulimazeni isithunzi somuntu siqu nesomsebenzi kanye nosongo lwemiphumela yezomthetho nezezimali. Kuze kwaphakanyiswa ukuthi ukusetshenziswa kabi noma ukungaphathwa kahle kwe-generative AI kungabeka ukuphepha kwezwe engcupheni.

Lezi zingozi azizange zibalekele osopolitiki. Ngo-Ephreli 2023, i-European Union yenza isiphakamiso imithetho emisha ye-copyright ye-AI ekhiqizayo okungadinga izinkampani ukuthi zidalule noma yiziphi izinto ezinelungelo lobunikazi ezisetshenziselwa ukuthuthukisa amathuluzi obuhlakani bokwenziwa akhiqizayo. Le mithetho yagunyazwa emthethweni osalungiswa ovotelwe yiPhalamende LaseYurophu ngoJuni, okuhlanganisa nemikhawulo eqinile ekusetshenzisweni kobuhlakani bokwenziwa emazweni angamalungu e-EU, okuhlanganisa nokuvinjelwa okuhlongozwayo kobuchwepheshe bokubona ubuso besikhathi sangempela ezindaweni zomphakathi.

Imisebenzi ezenzakalelayo nge-generative AI iphinde iphakamise ukukhathazeka mayelana nabasebenzi kanye nokususwa kwemisebenzi, njengoba kugqanyiswe uMcKinsey. Ngokusho kweqembu labaxhumanisi, i-automation ingabangela izinguquko zemisebenzi eyizigidi ezingu-12 phakathi kwamanje no-2030, nokulahlekelwa yimisebenzi kugxile ekusekelweni kwamahhovisi, ekuhlinzekeni amakhasimende kanye nasekuhlinzekeni ukudla. Lo mbiko ulinganisela ukuthi isidingo sabasebenzi basemahhovisi singase “… sehle ngemisebenzi eyizigidi eziyi-1,6, ngaphezu kokulahlekelwa yizi-830.000 zabathengisi bezitolo, ezingama-710.000 zabasizi bokuphatha kanye nezingu-630.000 zabathatha imali.”

I-Generative AI kanye ne-AI ejwayelekile

I-Generative AI kanye ne-AI evamile imelela izinhlangothi ezihlukene zohlamvu lwemali olufanayo. Kokubili kuthinta inkambu yobuhlakani bokwenziwa, kodwa eyokuqala iwuhlobo oluncane lwakamuva.

I-Generative AI isebenzisa amasu okufunda omshini ahlukahlukene, njenge-GAN, i-VAE, noma i-LLM, ukuze ikhiqize okuqukethwe okusha kumamodeli afundwe kudatha yokuqeqeshwa. Le miphumela ingaba umbhalo, izithombe, umculo, nanoma yini enye engamelwa ngokwedijithali.

I-Artificial general intelligence, eyaziwa nangokuthi i-artificial general intelligence, ibhekisela kabanzi kumqondo wezinhlelo zamakhompiyutha namarobhothi anobuhlakani obunjengobomuntu kanye nokuzimela. Lokhu kuseyindaba eqanjiwe yesayensi: cabanga i-WALL-E ka-Disney Pixar, uSonny ka-2004 I, Robot, noma i-HAL 9000, ubuhlakani bokwenziwa obunonya obuvela ku-Stanley Kubrick ka-2001: A Space Odyssey. Amasistimu amaningi amanje e-AI ayizibonelo “ze-AI encane”, njengoba yakhelwe imisebenzi eqondile.

I-Generative AI nokufunda komshini

Njengoba kuchazwe ngenhla, i-AI ekhiqizayo iyindawo engaphansi yobuhlakani bokwenziwa. Amamodeli e-Generative AI asebenzisa amasu okufunda omshini ukuze acubungule futhi akhiqize idatha. Ngokuvamile, ubuhlakani bokwenziwa bubhekisela kumqondo wamakhompyutha akwazi ukwenza imisebenzi ebingadinga ubuhlakani bomuntu, njengokuthatha izinqumo kanye ne-NLP.

Ukufunda ngomshini kuyisici esibalulekile sobuhlakani bokwenziwa futhi kubhekisela ekusetshenzisweni kwama-algorithms ekhompuyutha kudatha ngenjongo yokufundisa ikhompuyutha ukwenza umsebenzi othile. Ukufunda ngomshini kuyinqubo evumela amasistimu obuhlakani bokwenziwa ukuthi enze izinqumo ezinolwazi noma izibikezelo ngokusekelwe kumaphethini afundiwe.

Ingabe ubuhlakani bokwenziwa bokukhiqiza buyikusasa?

Ukukhula okunamandla kwe-AI ekhiqizayo akubonisi zimpawu zokuncipha, futhi njengoba izinkampani eziningi zamukela ukwenziwa kwedijithali kanye ne-automation, i-generative AI ibukeka izodlala indima ebalulekile esikhathini esizayo semboni. Amakhono okukhiqiza i-AI asevele afakazela ukuthi abalulekile ezimbonini ezifana nokudala okuqukethwe, ukuthuthukiswa kwesofthiwe, kanye nemithi, futhi njengoba ubuchwepheshe buqhubeka nokuvela, ukusetshenziswa kwabo kanye namacala okusebenzisa kuzokhula.

Sesikushilo lokho, umthelela we-AI ekhiqizayo kumabhizinisi, abantu ngabanye kanye nomphakathi uwonke uncike ekutheni sibhekana kanjani nezingozi eziba nazo. Ukuqinisekisa ukuthi i-artificial intelligence iyasetshenziswa ngokuziphatha ukunciphisa ukuchema, ukwenza ngcono ukuvuleleka nokuziphendulela kanye nokweseka ukubusa yedatha izobaluleka, kuyilapho uqinisekisa ukuthi ukulawulwa kuhambisana nokuthuthuka okusheshayo kobuchwepheshe kakade kubonakala kuyinselele. Ngokufanayo, ukuthola ibhalansi phakathi kokuzenzakalela kanye nokubandakanyeka komuntu kuzobaluleka uma sithemba ukusebenzisa amandla aphelele e-AI ekhiqizayo kuyilapho sinciphisa noma yimiphi imiphumela emibi.

Ercole Palmeri