- Ukucabanga okujulile kuhlanganisa amamodeli olimi athuthukile namaketanga okucabanga angaphakathi ukuxazulula izinkinga eziyinkimbinkimbi isinyathelo ngesinyathelo.
- Amathuluzi afana ne-Copilot Studio asebenzisa la mamodeli kuphela emisebenzini ebalulekile esebenzisa amagama angukhiye, ukulinganisela ukunemba, izindleko, kanye nesivinini.
- Ukufunda okujulile kanye nezakhiwo ezifana nama-CNN, ama-ViT, kanye nama-transformer kubeka isisekelo sobuchwepheshe sezinhlelo zokusebenza zezokwelapha, ezezimali kanye nezinsizakalo zamakhasimende.
- Ukwamukelwa kwayo kudinga ukuhlola izizathu, ukunemba, kanye ne-AI enomthwalo wemfanelo, ukunciphisa imikhawulo efana nokubambezeleka kanye nengozi yezimpendulo ezingalawulwa kahle.
Ubuhlakani bokwenziwa buthatha igxathu elikhulu phambili ngenxa yesizukulwane esisha samamodeli anekhono lokucabanga okuhlelekile kakhulu. Abakhi nje umbhalo noma izithombe: bangahlukanisa izinkinga, bahlole izinketho, futhi bathethelele izimpendulo zabo isinyathelo ngesinyathelo. Yilokhu, ezweni lobuchwepheshe, okubizwa ngokuthi ukucabanga okujulile.
Ukuqonda kahle ukuthi ukucabanga okujulile kubuhlakani bokwenziwa kuyini , ukuthi kuhluke kanjani ekufundeni okujulile kwakudala, nokuthi kusetshenziswa kanjani kumathuluzi angempela njengeCopilot Studio noma emikhakheni efana nezokwelapha noma isevisi yamakhasimende kubalulekile ekuqondeni ukuthi i-AI iya kuphi. Sizochaza konke lokhu ngokuthula, kodwa sisebenzisa ulimi olucacile nolufinyeleleka kalula ngangokunokwenzeka.
Kuyini ukucabanga okujulile ekuhlakanipheni okwenziwe?
Uma sikhuluma ngokucabanga okujulile, sibhekisela kumamodeli olimi athuthukile kakhulu, amakhulu aklanyelwe ngqo ukusingatha imisebenzi eyinkimbinkimbi edinga izinyathelo eziningi zokucabanga. Esikhundleni sokunikeza impendulo eqondile nengacacile, zithatha isikhathi "ukucabanga ngaphakathi," zikhiqiza uchungechunge lwemicabango yangaphakathi ngaphambi kokwethula umphumela kumsebenzisi.
Lawa mamodeli ayakwazi ukucabanga okunengqondo, ukuhlaziya okuningiliziwe, kanye nokuxazulula izinkinga ngokuhlukanisa umsebenzi ube yizinkinga ezincane. Nakuba le nqubo yokucabanga yangaphakathi ingafundiswa njalo kumsebenzisi wokugcina, ikhona futhi yilokho okubavumela ukuthi banikeze izimpendulo eziyinkimbinkimbi kakhulu kunezindlela zolimi zendabuko.
Umqondo oyinhloko ukuthi imodeli yokucabanga ejulile ayigcini nje ekubikezeleni igama elilandelayo ngokusekelwe kuzibalo kuphela; izama ukulandela ukulandelana kweziphetho zangaphakathi, okufana nendlela umuntu angaxazulula ngayo inkinga yezibalo, ahlaziye idatha yezezimali, noma athuthukise isu.
Empeleni, lokhu kuvula ithuba lokuthi i-ejenti ye-AI ibhekane nemisebenzi emide, ehlukahlukene eyayicishe ibe yindawo yochwepheshe babantu kuphela: kusukela ekuhlaziyeni izitayela zemakethe kuya ekuphawuleni embikweni wezokwelapha oyinkimbinkimbi.
Ukucabanga okujulile ku-Copilot Studio: ukuthi isebenza kanjani nokuthi isetshenziswa kanjani
Lapho eklama i-ejenti, umdali uchaza imiyalelo nemisebenzi okumele ayenze . Le misebenzi ingasukela ekusebenzisaneni okulula kakhulu (ukuphendula imibuzo evame ukubuzwa) kuya ekugelezeni okuyinkimbinkimbi kakhulu okudinga ukuhlaziywa okuphelele. Ezinyathelweni ezidinga ukuhlaziywa okujulile, imodeli ingayalwa ukuba isebenzise ukucabanga okuthuthukisiwe.
Ukuze usebenzise lawa makhono ku-Copilot Studio, engeza igama elingukhiye elithile (“isizathu”) ezinyathelweni ezifanele emiyalweni ye-ejenti. Isibonelo: “Sebenzisa isizathu ukuthola inombolo elandelayo ochungechungeni 2, 5, 10, 17…”. Ngaleso sikhathi, ngesikhathi sokwenziwa, i-Copilot ibiza imodeli yokucabanga ejulile (njengamanje eyimodeli ye-Azure OpenAI o3) ukucubungula leyo ngxenye ethile.
Le ndlela yesinyathelo ngesinyathelo ivumela unjiniyela ukuthi alawule ukuthi umenzeli wenza nini ukucabanga okuthuthukisiwe nokuthi impendulo esheshayo nelula yanele. Lokhu kuthuthukisa ibhalansi phakathi kwekhwalithi yokuphendula, izindleko zokubala, kanye nesivinini.
Ngamafuphi, i-Copilot Studio isebenzisa ukucabanga okujulile njengokuqinisa okuqondiwe kwamakhono okwenza izinqumo e-ejenti , hhayi njengento ehlala inikwe amandla kukho konke. Lokhu kuyenza isebenze ezimweni zebhizinisi langempela.
Ukusetshenziswa okuhlosiwe kokucabanga okujulile: kusukela kwezezimali kuya kwizibalo ezithuthukisiwe
Amamodeli okucabanga okujulile asebenza kahle emisebenzini eyinkimbinkimbi enezinyathelo eziningi kanye nokuncika . Ezinye izibonelo ezicacile zokusetshenziswa kwawo, esezivele zisetshenziswa kumapulatifomu afana neCopilot Studio, yilezi ezilandelayo:
Ukuhlaziywa kwemikhuba yemakethe kanye nezincomo zokutshalwa kwezimali : imodeli ingahlukanisa idatha yezezimali ibe yizicucu ezilawulekayo, uchungechunge lwesikhathi sokutadisha, ulwazi lomlando oluhlanganisa izinkomba, izimo zamanje zemakethe kanye nokubikezela, bese kusukela lapho incoma amathuba okutshalwa kwezimali athembisayo kakhulu.
Ukuphathwa kwempahla kanye nokubikezela isidingo : Ngokusekelwe olwazini mayelana nokuthengiswa kwangaphambilini, isikhathi sonyaka, ukugeleza kochungechunge lokunikezwa kanye nezinguquko ekuziphatheni kwamakhasimende, imodeli ingaphakamisa amasu esitoko , icabange kabusha amazinga okuphepha futhi iphakamise ukulungiswa ukuze kuncishiswe ukuphela kwesitoko noma izimpahla eziningi ngokweqile.
Ukuxazulula izilinganiso ezihlukile kanye nezinkinga zezibalo eziyinkimbinkimbi : lawa mamodeli angabhekana nezivivinyo ezithuthukisiwe ngokuhlukanisa inkinga ibe yizinyathelo ezinengqondo ezixhumene , echaza okwenziwayo esigabeni ngasinye, into ewusizo kakhulu ezimweni zemfundo noma zocwaningo.
Empeleni, noma nini lapho inqubo yokucabanga engeyona ewubala idinga ukulandelwa , amamodeli okucabanga ajulile anikeza inani elikhulu kakhulu uma kuqhathaniswa namamodeli alula aphendula ngokuqondile kuphela.
Ukuhlola, izilinganiso, kanye nokuziphendulela kumamodeli okucabanga okujulile
Ukuze lezi zinhlelo zibe usizo ezindaweni zangempela, akwanele ukuthi zibe "zihlakaniphile": kumele zithembeke, zivikeleke, futhi zilandeleleke . Yingakho amamodeli okucabanga ajulile njengalawo asetshenziswa ku-Copilot Studio ehlolwa kaningana ngaphambi kokuba atholakale kubasebenzisi.
Okokuqala, isisekelo semodeli siyahlolwa ; okungukuthi, ukuqinisekisa ukuthi isekelwe kumongo wezwe langempela futhi ayimane nje isungule idatha ngokungahleliwe. Lokhu kwenziwa ngokuhlola imodeli ezimweni ngolwazi olwaziwayo nokuhlola ukuthi ihlala ithembekile kangakanani kulowo mongo.
Okwesibili, ukuhambisana nezimiso ze -AI ezithembekile kuyahlaziywa : ukuvikelwa emizamweni yokuqhekeka kwejele (ukuphoqa imodeli ukuthi idlule imingcele yayo), ukuhlaselwa kokujova kwesizinda esihlukile (ukuxubana okunonya kwemiyalelo evela emithonjeni ehlukene) kanye nokuhlunga okuqukethwe okulimazayo noma okungafanele.
Okokugcina, ukunemba kwezimpendulo kulinganiswa kuzo zonke izimo zokusetshenziswa eziningi. Ukusebenza kunikezwa amaphuzu kuzo zonke lezi zilinganiso kusethi ehlukahlukene yokuhlolwa, ngakho-ke amamodeli ahlangabezana nezinga elithile kuphela ashicilelwa ekugcineni.
Lolu hlobo lokuhlola okuqhubekayo luyisihluthulelo sokuqinisekisa ukuthi amamodeli okucabanga okujulile awabi "ibhokisi elimnyama elingalawuleki ," kodwa kunalokho abe amathuluzi izinhlangano ezingawahlola, ziwahlole, futhi ziwaqhathanise.
Ukulinganiselwa kokucabanga okujulile kanye nendlela yokukunciphisa
Ukuthi nje imodeli "ijule kakhulu" akusho ukuthi iphelele. Eqinisweni, lezi zinhlelo zinemikhawulo ebalulekile , njengemisebenzi lapho i-ChatGPT yehluleka khona , okubalulekile ukuyiqonda.
Esokuqala isikhathi sokuphendula . Ngenxa yokuthi zidinga ukubalwa kwangaphakathi okwengeziwe, amamodeli okucabanga ngokuvamile athatha isikhathi eside kunamamodeli olimi ajwayelekile. Lokhu kungaba yinkinga ezinhlelweni zokusebenza ezizwela ukubambezeleka (isibonelo, ingxoxo nekhasimende elingenasineke) uma zisetshenziswa ngokweqile.
Umkhawulo wesibili ukuthi i-ejenti ingasebenzisa lezi zinhlobo zamamodeli kuphela uma ukucushwa kwayo kukuvumela . Ngamanye amazwi, imisebenzi yokucabanga okujulile kumele isebenze ngokucacile ku-ejenti; ayisebenzi ngokuzenzakalelayo kuzo zonke izimo.
Ukuze unciphise lezi zingqinamba, kunconywa ukuthi: usebenzise ukucabanga okujulile kuphela kuma-ejenti akudingayo ; sebenzisa igama elingukhiye elihambisanayo kuphela ezinyathelweni ezizuza ngempela ekuhlaziyeni okujulile; futhi ugcine la mamodeli emisebenzini lapho kwamukelekile ukulinda isikhathi eside ukuze uthole ikhwalithi yokuphendula ephezulu.
Ngaphezu kwalokho, kuwumkhuba omuhle ukuxwayisa abasebenzisi bokugcina ukuthi ezinye izimpendulo eziyinkimbinkimbi zama-ejenti zingathatha imizuzwana embalwa eyengeziwe, ukuze kulawulwe okulindelwe futhi kugwenywe ukukhungatheka.
Imikhuba emihle yokusebenza yokusetshenziswa okunesibopho
Ngokombono wokusebenza, kunezindlela eziningana zokuqinisekisa ukuthi ukucabanga okujulile kusetshenziswa ngempumelelo nangokuphephile ezinhlanganweni.
Okokuqala ukukhawulela ukusetshenziswa kwayo kuma- ejenti adinga ngempela uchungechunge lokucabanga oluyinkimbinkimbi , njengokuhlaziywa kwedatha okungahlelekile, izinqumo ezibucayi ezisekelwe ezicini eziningi, noma ukwenziwa kwemibiko emide nefanele. Ukuyivumela ngokuzenzakalelayo kukho konke kuwukuchitha izinsiza.
Isincomo sesibili ukuhlola ngokuphelele i-ejenti, ukubuyekeza ikhwalithi, ukuvumelana, kanye nokuthembeka kwemiphumela lapho imodeli yokucabanga isetshenziswa. Lezi zivivinyo zikuvumela ukuthi uthole izimo lapho imodeli ingase yehluleke, idale idatha, noma ingavumelani kahle nesizinda esithile senkampani.
Amathuluzi afana nemephu yomsebenzi akuvumela ukuthi ubone ukuthi, ngesikhathi seseshini, umenzeli uhlanganyele nini ekucabangeni okujulile, abukeze izinyathelo zangaphakathi, futhi aqhathanise imiphumela. Lokhu kusiza ekunqumeni ukuthi imodeli iletha ngempela inani elilindelekile.
Kuwusizo futhi ukuqhathanisa imiphumela ngokuhlelekile kanye nangaphandle kokucabanga okujulile ngokubuyekeza imiyalelo kanye nokusebenzisa izivivinyo ze-A/B. Lokhu kukhomba ukuthi yiziphi izingxenye zokugeleza ezizuza ngokusebenzisa lawa mamodeli nokuthi imodeli yolimi elula yanele kuphi.
Kusukela ekufundeni komshini kuya ekufundeni okujulile kanye ne-AI yokukhiqiza
Ukuze kubekwe kahle ukucabanga okujulile, kuyasiza ukubuyekeza ukuvela kobuchwepheshe be-AI: ukufunda komshini, ukufunda okujulile, kanye ne-AI yokukhiqiza kwakha uhlobo lwesitebhisi sobunzima.
Ukufunda komshini kwendabuko kwakuncike kakhulu ekufundeni okuqondisiwe. Isibonelo, ukudala uhlelo olungabona izilwane ezithombeni, kwakudingeka ukulebula ngesandla amakhulu ezinkulungwane zezithombe, ukuqeqesha i-algorithm, ukuyihlola ngezithombe ezintsha, ukuhlaziya amaphutha, bese wandisa isethi yedatha enelebula futhi ukuze kuthuthukiswe ukunemba.
Le nqubo yadinga ukungenelela okukhulu komuntu kulokho okubizwa ngokuthi ubunjiniyela bezimfanelo : ukunquma ukuthi yiziphi izici okufanele zikhishwe (umbala, ukuthungwa, izimo, njll.) nokuthi uzimele kanjani ngezinombolo ukuze i-algorithm ifunde.
Ngokufunda okujulile , amanethiwekhi e-neural anezingqimba eziningi athatha umsebenzi wokufunda ngqo lezi zimelelwano kusuka kudatha eluhlaza (izithombe, umbhalo, umsindo, njll.). Kwakungasadingeki ukuchaza ngesandla zonke izimfanelo: imodeli ngokwayo yathola ukuthi yimaphi amaphethini afanele.
Isinyathelo esilandelayo amamodeli e-AI akhiqizayo asekelwe kuzakhiwo eziguqulayo . Lawa mamodeli, njengamamodeli ezilimi ezinkulu, awagcini nje ngokuqaphela amaphethini kodwa futhi akhiqiza okuqukethwe okusha (umbhalo, izithombe, ikhodi) ngokuhlanganisa lawo maphethini ngezindlela ezintsha.
Ukucabanga okujulile kwakhelwe kulesi sisekelo: kuhlanganisa amandla okukhiqiza ama-transformer namasu angaphakathi aklanyelwe ukulandela uchungechunge lwemicabango olude noluhlelekile.
Izinzuzo zokufunda okujulile ngaphezu kokufunda komshini kwakudala
Ukufunda okujulile kwethule izinzuzo eziningi ezicacile kune-ML yendabuko , okuyisisekelo sobuchwepheshe sentuthuko eminingi ekucabangeni okujulile.
Ngakolunye uhlangothi, inikeza ukucubungula okuphumelelayo kakhulu kwedatha engahlelekile (umbhalo wamahhala, izithombe, umsindo). Ngenkathi i-algorithm yakudala inganqotshwa yizindlela eziningi ezingapheli zokuveza umqondo ofanayo, inethiwekhi ejulile ingafunda ngokuqondile lokhu kufana. Ngakho-ke, ingaqonda ukuthi "Ngingayidlulisela kanjani imali?" kanye "Ungachaza ukuthi ungayenza kanjani inkokhelo?" kubhekisela ohlotsheni olufanayo lwesenzo.
Ngaphezu kwalokho, amanethiwekhi ajulile amahle kakhulu ekudaluleni ubudlelwano obufihliwe kanye namaphethini angalindelekile . Imodeli eqeqeshwe ngedatha yokuthenga ingaphakamisa imikhiqizo ikhasimende elingakayithengi, ngokuqhathanisa ukuziphatha kwalo nokwabasebenzisi abafanayo, noma ngabe abakafundiswanga ngokucacile leso sincomo esithile.
Enye inzuzo yikhono lokufunda ungagadiwe noma ungagadiwe kancane . Amanethiwekhi angazivumelanisa nokuziphatha komsebenzisi ngokuhamba kwesikhathi ngaphandle kokudinga izigidi zamaphuzu edatha anelebula. Isihloli sokupela esizenzakalelayo, isibonelo, singahlanganisa amagama kwezinye izilimi njengoba umsebenzisi evame ukuwathayipha.
Okokugcina, ukufunda okujulile kunamandla kakhulu ngedatha eguquguqukayo noma eguquguqukayo kakhulu , njengokuthengiselana kwezezimali. Kungafunda ukuhlukanisa amaphethini ajwayelekile okuziphatha kokukhokha futhi kubeke uphawu kulawo ahluka njengokukhwabanisa okungenzeka.
Ukusetshenziswa kokufunda okujulile: kusukela kwezokwelapha kuya ezimotweni ezizishayelayo
Ukufunda okujulile sekusabalele kakhulu kwezokwelapha nasekuxilongeni . Kusetshenziselwa ukuhlukanisa izithombe zezokwelapha, izingxenye zezitho kanye nezilonda, ukuhlaziya i-histopathology yedijithali, kanye nokusiza ekuxilongeni kusetshenziswa ama-X-ray, ama-MRI, kanye nokuhlolwa kwelebhu.
Kusukela ngo-2022, izakhiwo ze -Vision Transformer (ViT) zibonise ukusebenza okufana noma okuphakeme kakhulu kunezinethiwekhi ze-neural ze-convolutional zakudala emisebenzini emikhulu yokuhlukanisa izithombe zezokwelapha. Inzuzo yazo eyinhloko ikhono lazo lokuthwebula ukuncika komhlaba wonke ezithombeni ze-gigapixel ngokusebenzisa izindlela zokuzinakekela ezisezingeni eliphezulu.
Ngaphandle kwensimu yezokwelapha, ukufunda okujulile kungemuva kwezimoto ezizisebenzela zona, ukuqashelwa kobuso, abasizi bezwi abanjengo-Alexa no-Siri, izinhlelo ze-TV nezomculo, nokunye okuningi. Kuzo zonke lezi zimo, imodeli kumele ichaze idatha yangempela enomsindo neguquguqukayo kakhulu.
Ukuhlolwa okuphawulekayo okufana ne -AlphaGo , okwafunda ukudlala i-Go nokunqoba ochwepheshe abaphezulu babantu, kubonise izinga lapho inethiwekhi yezinzwa ezijulile ingafinyelela khona amakhono abhekwa “njengokuzibonela” noma “okudala,” ngaphandle kokuthi umhleli wezinhlelo atshele yonke into eyenzeka.
Kuyini ngempela ukufunda okujulile: izendlalelo, amazinga, kanye namandla ekhompyutha
Nakuba kungekho ncazelo eyodwa ebanzi, iningi labacwaningi liyavuma ukuthi ukufunda okujulile kusekelwe ezingqimbeni eziningi zokucubungula okungeyona imigqa ezikhipha izici ezingaqondakali ezivela kudatha.
Ezingqimbeni ezingezansi, kufundwa izici ezilula (imingcele esithombeni, inhlanganisela yamagama ayisisekelo), kuyilapho izingqimba ezingaphezulu zihlanganisa lezo zici ukuze zakhe imiqondo eyinkimbinkimbi kakhulu (ubuso, izinto, izincazelo zemisho).
Umehluko ngama-algorithms "angajulile" ulele kakhulu enanini lokuguqulwa okuboshwe ngeketanga . Ngenkathi imodeli yakudala ingasebenzisa ukuguqulwa okukodwa noma okubili, imodeli ejulile ingaba namashumi noma amakhulu ezendlalelo eziphakathi, okuvumela ukuthi imele imisebenzi eyinkimbinkimbi kakhulu.
Izindleko ukuthi ukuqeqesha amanethiwekhi ajulile kudinga amandla amaningi okusebenzisa ikhompyutha . Yingakho ama-GPU eseyithuluzi eliyisisekelo lokuqeqesha la mamodeli, ngenxa yekhono lawo lokwenza imisebenzi emikhulu efana (i-GPGPU).
Abahlinzeki abakhulu bamafu (i-Amazon, i-Azure, i-IBM, i-Google, njll.) sebevele banikeza ingqalasizinda ngama-GPU akhethekile kanye namapulatifomu e-PaaS okufunda komshini, njengalawo asekelwe ku-TensorFlow, anamamodeli namathuluzi aqeqeshwe kusengaphambili ukuze azivumelanise nesimo ngasinye.
Ama-algorithms okufunda okujulile kanye nezakhiwo ezifanele kakhulu
Ngaphakathi kokufunda okujulile, kuye kwavela izinhlobo eziningi zamanethiwekhi e-neural , ngalinye lilungiselelwe uhlobo oluthile lwedatha noma inkinga.
Amanethiwekhi e-Convolutional neural (ama-CNN) aklanyelwe ukucubungula izithombe namavidiyo. Asebenzisa izihlungi (ama-convolution) eziskena isithombe ukuze zithole amaphethini endawo bese ziwahlanganisa. Ayisisekelo sokubona kwekhompyutha yesimanje: ukuqashelwa kobuso, ukuhlukaniswa kwezinto, ukuhlaziywa kwezithombe zezokwelapha, nokuningi.
Amanethiwekhi e-neural aphindaphindayo (ama-RNN) , kanye nezinhlobo zawo zesimanje, afaka ama-feedback loops avumela ukuthi "akhumbule" ulwazi oludlule. Lokhu kuwenza abe usizo kakhulu ezinhlotsheni, njengombhalo, umsindo, noma uchungechunge lwesikhathi. Isibonelo, uhlelo lokuzulazula lungasebenzisa lezi zinkumbulo ukulindela ukuminyana kwethrafikhi okuvamile futhi luphakamise izindlela ezihlukile.
Ngesikhathi esifanayo, imiqondo efana nokufunda ndawonye, amanethiwekhi asele, ama-vision transformers , kanye neminye eminingi iye yavela, ekhulisa futhi ihlanza ikhono lamanethiwekhi ajulile lokuzivumelanisa nezinkinga ezithile.
Lonke lolu hlelo lobuchwepheshe yilona olwenza kube nokwenzeka, namuhla, ukwakha amamodeli anezizathu ezijulile phezu kwawo: ngaphandle kwalesi sisekelo sezakhiwo ezinamandla nezikwaziyo ukwandiswa, bekungeke kwenzeke.
Izinhlobo zokucabanga ezingalingiswa yi-AI
Izinhlelo ze-AI zanamuhla zingahlanganisa amasu okucabanga ahlukahlukene ahlukahlukene , kuye ngohlobo lwedatha kanye nohlelo lokusebenza oluqondiwe. Azikhawulelwe "endleleni eyodwa yokucabanga."
Izindlela ezivamile zifaka phakathi ukucabanga okusekelwe emithethweni evamile kuze kufike eziphethweni ezithile, ukucabanga okusekelwe emibonweni (ukuhlanganisa izinto ngezibonelo), ukucabanga okungenzeka (ukusebenza ngokungaqiniseki), kanye nokucabanga okungacacile (ukusingatha amagama angacacile njengokuthi "okuphezulu", "okuphansi", "okuphakathi").
Izindlela ezifana nokucabanga okucashile (ukuphakamisa incazelo enengqondo kakhulu yeqoqo lamaqiniso), ukucabanga okunengqondo okuvamile , ukucabanga kwendawo kanye nesikhathi (okubaluleke kakhulu kumarobhothi kanye nokushayela okuzenzakalelayo), kanye nokucabanga okufana nokwe-neurosymbolic , okuhlanganisa amanethiwekhi ezinzwa nokuqonda okungokomfanekiso, nakho kuyahlolwa.
Ukucabanga okujulile kuncike kuleli bhokisi lamathuluzi ukwakha uchungechunge olucebile lokucabanga , ukuxuba idatha, izibalo, kanye nemithetho uma kudingeka.
I-AI, ukufunda komshini nokufunda okujulile: umehluko obalulekile
Ukuze kucaciswe amagama: ubuhlakani bokwenziwa (AI) yigama elibanzi kakhulu, elihlanganisa noma yiluphi uhlelo olukwazi ukwenza imisebenzi ehlobene nobuhlakani bomuntu (ukucabanga, ukufunda, ukuqonda, njll.). Ngaphakathi kwe-AI, sinokufunda komshini (ML) , okugxila kuma-algorithms afunda kudatha ngaphandle kokuhlela ngokwesisekelo secala ngalinye.
Ukufunda okujulile , nakho, kuyisigaba esincane sokufunda komshini esisebenzisa amanethiwekhi e-neural anezingqimba eziningi ukuze afunde ngqo kusuka kudatha enkulu. Umehluko omkhulu usekwakhekeni kwemodeli kanye nendlela izici ezikhishwa ngayo.
Ezingeni elisebenzayo, i-ML yakudala ivame ukudinga umsebenzi owengeziwe wezandla kubunjiniyela bezici , idatha encane kanye namandla amancane ekhompyutha, kuyilapho ukufunda okujulile kudinga amasethi edatha amakhulu, ama-GPU anamandla kanye nezikhathi zokuqeqesha ezinde, kodwa kunikeza intuthuko emangalisayo kumakhono emisebenzi eyinkimbinkimbi kanye nedatha engahlelekile.
Ngokuqondene nokuhunyushwa kalula, amamodeli alula okufunda komshini (ukuhlehla okuqondile, izihlahla ezingajulile) kulula ukuwachaza, kuyilapho amanethiwekhi ajulile eziphatha njenge "black boxes." Lokhu kuthinta namamodeli okucabanga okujulile, azuza okunye kwalokhu kungabonakali, yize imizamo iyaqhubeka yokuwenza abe sobala kakhudlwana.
Ukucabanga okujulile kanye nenkonzo yamakhasimende
Enye yezinkambu lapho ukusetshenziswa okungokoqobo kwe-AI kanye nokufunda okujulile kukhula khona kakhulu yinkonzo yamakhasimende . Izinhlelo eziningi zamanje zisebenzisa ama-algorithms e-ML ukuze zizisebenzele zona, zisekele ama-ejenti abantu, kanye nokuhlelwa komsebenzi.
Idatha ephakela lezi zinhlelo ivela emibuzweni yangempela yamakhasimende , umlando wezigameko, umongo wokuthenga, kanye nokuziphatha kokusetshenziswa. Njengoba la mamodeli ephakelwa, izibikezelo neziphakamiso ziyashesha futhi zinembe kakhudlwana.
Kulesi simo, ukucabanga okujulile kuvumela ama-bot ukuthi angaphenduli imibuzo elula kuphela, kodwa futhi ahlaziye isimo sonke sekhasimende , abukeze umlando wawo, ahlole izixazululo ezahlukahlukene ezingaba khona futhi aphikisele esingcono kakhulu, ngezinga elikhulu lokwenza ngokwezifiso.
Amapulatifomu akhethekile, njenge-bots ethuthukisiwe yezixazululo ezithile ze-CX, asevele ehlanganisa izizindalwazi ezinkulu zenhloso yamakhasimende namamodeli okufunda okujulile ukuze anikeze izimpendulo zemvelo neziwusizo kakhudlwana , okwandisa umkhiqizo wama-ejenti abantu futhi kunciphise ukusethwa kokugeleza kokusekela.
Njengoba amamodeli okucabanga ajulile ehlanganiswa kangcono nalezi zinhlobo zamathuluzi, sizobona ama-ejenti abonakalayo ekwazi ukubamba izingxoxo ezinde neziyinkimbinkimbi , ukugcina intambo, ukusekela izinqumo, kanye nokuzivumelanisa nethoni yomsebenzisi cishe njengoba umuntu engenza.
Lonke lolu hambo, kusukela ekufundeni komshini okuvamile kuya ekufundeni okujulile, i-AI yokukhiqiza, kanye nokucabanga okujulile, kwembula indlela ecacile: sisondela kakhulu ezinhlelweni ezingagcini nje ngokuqaphela amaphethini kodwa futhi ezikwazi ukucabanga okuhlelekile ngezinkinga eziyinkimbinkimbi . Inselelo manje akuyona nje eyobuchwepheshe kodwa futhi eyokuziphatha nokusebenza: ukuqinisekisa ukuthi lawa mamodeli ahlolwa kahle, asetshenziswa lapho enikeza khona inani langempela, izingozi zawo ziyaphathwa, futhi ahlanganiswe ngokuzibophezela kumathuluzi afana ne-Copilot Studio, ukuze ubuhlakani bokwenziwa bube umlingani onamandla nothembekile emsebenzini wansuku zonke.
