Uyini umthwalo wemfanelo wabathuthukisi abasebenzisa i-Generative AI? [Ividiyo Nombuzo]

Impendulo emfushane: Onjiniyela abasebenzisa i-AI ekhiqizayo banesibopho sohlelo lonke, hhayi nje kuphela umphumela wemodeli. Uma i-AI ithonya izinqumo, ikhodi, ubumfihlo, noma ukwethenjwa komsebenzisi, kumele bakhethe izinhlelo zokusebenza eziphephile, baqinisekise imiphumela, bavikele idatha, banciphise umonakalo, futhi baqinisekise ukuthi abantu bangabuyekeza, bashintshe, futhi balungise amaphutha.

Izinto ezibalulekile okufanele uzicabangele:

Ukuqinisekiswa: Phatha imiphumela ephucukisiwe njengengathembekile kuze kube yilapho imithombo, izivivinyo, noma ukubuyekezwa komuntu kuziqinisekisa.

Ukuvikelwa kwedatha: Nciphisa idatha esheshayo, susa izihlonzi, futhi uvikele amalogi, izilawuli zokufinyelela, kanye nabathengisi.

Ubulungiswa: Hlola kuzo zonke izinhlobo zabantu kanye nezimo ukuze uthole imibono engafani kanye namaphethini okwehluleka angalingani.

Ukucaca: Faka ilebula ngokucacile ukusetshenziswa kwe-AI, chaza imikhawulo yayo, bese unikeza isibuyekezo noma isikhalazo somuntu.

Ukuziphendulela: Nika abanikazi abacacile bokuthunyelwa, izehlakalo, ukuqapha, kanye nokubuyiselwa emuva ngaphambi kokuqaliswa.

Izihloko ongase uthande ukuzifunda ngemva kwalesi:

🔗 Amathuluzi e-AI amahle kakhulu onjiniyela besofthiwe: Abasizi abaphezulu bokubhala amakhodi abasebenzisa i-AI
Qhathanisa abasizi abaphezulu bokubhala amakhodi e-AI ukuze uthole imisebenzi yokuthuthukisa esheshayo nehlanzekile.

🔗 Amathuluzi ayi-10 aphezulu e-AI onjiniyela ukuze bathuthukise umkhiqizo
Uhlu olubekwe ezikhundleni lwamathuluzi e-AI onjiniyela wokubhala ikhodi okuhlakaniphile kanye nesivinini.

🔗 Kungani i-AI ingaba yimbi emphakathini kanye nokwethembana
Ichaza umonakalo wangempela: ubandlululo, ubumfihlo, imisebenzi, kanye nezingozi zolwazi olungalungile.

🔗 Ingabe i-AI ihambe kakhulu ezinqumweni ezibaluleke kakhulu?
Ichaza ukuthi i-AI idlula nini imingcele: ukuqapha, ama-deepfake, ukukholisa, ukungavumi.

Kungani umthwalo wemfanelo wabathuthukisi abasebenzisa i-Generated AI ubaluleke kakhulu kunalokho abantu abakucabangayo

Iziphazamiso eziningi zesofthiwe ziyacasula. Inkinobho iyaphuka. Ikhasi lilayisha kancane. Kukhona okuphahlazekayo futhi wonke umuntu uyabubula.

Izinkinga ze-AI ezikhiqizayo zingahluka. Zingaba zicashile.

Imodeli ingazwakala iqiniseka ngenkathi ingalungile. Iphrofayela ye-NIST GenAI Ingaveza ubandlululo ngaphandle kwezimpawu ezisobala zesixwayiso. Iphrofayela ye-NIST GenAI Ingadalula idatha ebucayi uma isetshenziswa ngokunganaki. I-OWASP Top 10 yezicelo ze-LLM Imibuzo eyisishiyagalombili ye-ICO ye-AI yokukhiqiza Ingakhiqiza ikhodi esebenza - ize yehluleke ekukhiqizeni ngendlela ehlazisayo kakhulu. I-OWASP Top 10 yezicelo ze-LLM Kufana nokuqasha umfundi okhuthele kakhulu ongalali futhi ngezikhathi ezithile asungula amaqiniso ngokuzethemba okumangalisayo.

Yingakho umthwalo wemfanelo wabathuthukisi abasebenzisa i-Generative AI mkhulu kunokusetshenziswa okulula. Abathuthukisi abasakhi nje kuphela izinhlelo zokucabanga. Bakha izinhlelo ezingenzeka ezinemiphetho engacacile, imiphumela engalindelekile, kanye nemiphumela yangempela yomphakathi. I-NIST AI RMF

Lokho kusho ukuthi umthwalo wemfanelo uhlanganisa:

• ukuqonda imikhawulo yemodeli ye- NIST AI RMF

• ukuvikela ubumfihlo bomsebenzisi isiqondiso se-ICO mayelana ne-AI nokuvikelwa kwedatha

• ukunciphisa imiphumela eyingozi Iphrofayili ye-NIST GenAI

• ukuhlola ukunemba ngaphambi kokuba ukwethenjwa kunikezwe iphrofayela ye-NIST GenAI

• ukwenza indima yomuntu icace Izimiso ze-AI ze-OECD

• ukuklama izindlela zokubuyela emuva lapho i-AI yehluleka Izimiso ze-OECD AI Iziqondiso ze-NCSC ezivikelekile ze-AI

• ukubhala phansi uhlelo ngokucacile Izimiso ze-OECD AI

Uyazi ukuthi kuhamba kanjani - uma ithuluzi lizwakala limangalisa, abantu bayayeka ukulibuza. Abathuthukisi abakwazi ukubekezelela ukukhululeka okunjalo.

Yini eyenza inguqulo enhle yomthwalo wemfanelo wabathuthukisi abasebenzisa i-Generative AI? 🛠️

Inguqulo enhle yomthwalo wemfanelo ayisebenzi kahle. Akukhona nje ukwengeza isixwayiso phansi bese uyibiza ngokuthi ukuziphatha. Ibonakala ekukhetheni umklamo, emikhubeni yokuhlola, kanye nokuziphatha komkhiqizo.

Nansi indlela inguqulo eqinile yomthwalo wemfanelo wabathuthukisi abasebenzisa i-Generative AI evame ukubonakala ngayo:

• Ukusetshenziswa okuhlosiwe kwe -NIST AI RMF

  • I-AI isetshenziselwa inkinga yangempela, hhayi ukuphoqelelwa emkhiqizweni ngoba izwakala imfashini.

• Izimiso ze-AI ze-OECD zokuqondisa abantu

  • Abantu bangabuyekeza, balungise, bashintshe, noma benqabe imiphumela.

• Iziqondiso ze-AI ezivikelekile ze-NCSC ngokuklama

  • Izilawuli zezingozi zakhiwe ekuqaleni, azifakwanga ipayipi kamuva.

• Ukucaca Kwezimiso ze-OECD AI Ukubuka konke koMthetho we-European Commission AI

  • Abasebenzisi bayaqonda ukuthi okuqukethwe kukhiqizwa yi-AI noma kusizakala nge-AI.

• yokunakekelwa kwedatha ye-AI ekhiqizayo

  • Ulwazi olubucayi luphathwa ngokucophelela, futhi ukufinyelela kulinganiselwe.

• Ubulungisa buhlola -NIST GenAI Profile ICO mayelana ne-AI nokuvikelwa kwedatha

  • Uhlelo luhlolwa ukucwasa, ukusebenza okungalingani, kanye namaphethini ayingozi.

• Ukuqapha okuqhubekayo kwe-NIST AI RMF NCSC iziqondiso ze-AI eziphephile

  • Ukuqalisa akuyona intambo yokugcina. Kufana kakhulu nempempe yokuqala.

Uma lokho kuzwakala sengathi kuningi, kulungile... kunjalo. Kodwa yilokho okushiwoyo uma usebenza nobuchwepheshe obungathonya izinqumo, izinkolelo, kanye nokuziphatha ngezinga elithile. Izimiso ze-OECD AI

Ithebula Lokuqhathanisa - umthwalo wemfanelo oyinhloko wabathuthukisi abasebenzisa i-Generative AI ngokushesha 📋

Itafula elingalingani kancane, yebo, kodwa elifanelana nesihloko. Umthwalo wemfanelo wangempela nawo awulingani.

Umthwalo wemfanelo uqala ngaphambi kwesicelo sokuqala - ukukhetha indlela efanele yokusebenzisa 🎯

Omunye wemithwalo yemfanelo emikhulu onjiniyela abanayo ukunquma ukuthi i-AI ekhiqizayo kufanele isetshenziswe yini nhlobo. I-NIST AI RMF

Lokho kuzwakala kusobala, kodwa kuyeqiwa ngaso sonke isikhathi. Amaqembu abona imodeli, ajabule, bese eqala ukuyiphoqa ukuthi ingene emisebenzini ezosingathwa kangcono yimithetho, usesho, noma i-logic evamile yesofthiwe. Akuzona zonke izinkinga ezidinga imodeli yolimi. Ezinye izinkinga zidinga isizindalwazi kanye nentambama ethule.

Ngaphambi kokwakha, abathuthukisi kufanele babuze:

• Ingabe umsebenzi uvulekile noma uqinisekile?

• Ingabe umphumela ongalungile ungadala umonakalo?

• Ingabe abasebenzisi badinga ubuhlakani, ukubikezela, ukufingqa, ukuzenzekela - noma isivinini nje?

• Ingabe abantu bazokwethemba ngokweqile umphumela? Iphrofayela ye-NIST GenAI

• Ingabe umuntu angayibuyekeza imiphumela ngendlela engokoqobo? Izimiso ze-OECD AI

• Kwenzekani uma imodeli ingalungile? Izimiso ze-OECD AI

Umthuthukisi onomthwalo wemfanelo akabuzi nje ukuthi, “Singayakha yini le?” Bayabuza ukuthi, “Ingabe lokhu kufanele kwakhiwe ngale ndlela?” I-NIST AI RMF

Lowo mbuzo ngokwawo uvimbela izinto eziningi ezingenangqondo.

Ukunemba kuwumthwalo wemfanelo, hhayi isici sebhonasi ✅

Masicacise - esinye sezingibe ezinkulu ku-AI yokukhiqiza ukuphambanisa ukukhuluma kahle neqiniso. Amamodeli avame ukukhiqiza izimpendulo ezizwakala zipholile, zihlelekile, futhi zikholisa ngokujulile. Okuhle, kuze kube yilapho okuqukethwe kungenangqondo okugoqwe ngokuzethemba. Iphrofayili ye-NIST GenAI

Ngakho-ke umthwalo wemfanelo wabathuthukisi abasebenzisa i-Generated AI uhlanganisa ukwakha ukuze kuqinisekiswe.

Lokho kusho ukuthi:

• kusetshenziswa ukubuyisa noma ukusekela lapho kungenzeka khona iphrofayela ye-NIST GenAI

• ukuhlukanisa okuqukethwe okukhiqizwe kumaqiniso aqinisekisiwe Izimiso ze-OECD AI

• ukwengeza imikhawulo yokuzethemba ngokucophelela i-NIST AI RMF

• ukudala imisebenzi yokubuyekeza yemiphumela ebaluleke kakhulu Izimiso ze-OECD AI

• ukuvimbela imodeli ukuthi ingathuthuki ezimweni ezibucayi iphrofayela ye-NIST GenAI

• izixwayiso zokuhlola ezizama ukuphula noma ukudukisa uhlelo I-OWASP Eziyi-10 Eziphezulu zezinhlelo zokusebenza ze-LLM

Lokhu kubaluleke kakhulu ezindaweni ezifana nalezi:

• ukunakekela impilo

• ezezimali

• imisebenzi yezomthetho

• imfundo

• ukwesekwa kwamakhasimende

• ukuzenzekela kwebhizinisi

• ukukhiqizwa kwekhodi

Ikhodi ekhiqizwe, isibonelo, ingabukeka icocekile ngenkathi ifihla amaphutha okuphepha noma amaphutha okucabanga. Unjiniyela oyikopisha ngokungazi lutho akasebenzi kahle - umane nje unikeza abanye abantu ithuba lokusebenzisa ingozi ngendlela enhle kakhulu. I-OWASP Top 10 ye-LLM Applications NCSC ku-AI kanye nokuphepha kwe-cyber

Imodeli ingasiza. Unjiniyela usengumnikazi womphumela. Izimiso ze-OECD AI

Ubumfihlo kanye nokuphathwa kwedatha akunakuxoxiswana ngakho 🔐

Nakhu lapho izinto ziba zimbi khona ngokushesha. Izinhlelo ze-AI ezikhiqizayo zivame ukuthembela ezikhuthazweni, amalogi, amafasitela omongo, izendlalelo zememori, ukuhlaziya, kanye nengqalasizinda yomuntu wesithathu. Lokho kudala amathuba amaningi okuthi idatha ebucayi ivuze, iqhubeke, noma isetshenziswe kabusha ngezindlela abasebenzisi abangazange bazilindele. Imibuzo eyisishiyagalombili ye-ICO ye-AI ekhiqizayo I-OWASP Eziyi-10 Eziphezulu zezinhlelo zokusebenza ze-LLM

Abathuthukisi banesibopho sokuvikela:

• ulwazi lomuntu siqu

• amarekhodi ezezimali

• imininingwane yezokwelapha

• idatha yangaphakathi yenkampani

• izimfihlo zokuhweba

• amathokheni okuqinisekisa

• ukuxhumana kwamakhasimende

Imikhuba yokwethembeka ihlanganisa:

• ukunciphisa ukuthi iyiphi idatha engena emibuzweni eyisishiyagalombili yemodeli ye-ICO ye-AI ekhiqizayo

• ukufihla noma ukususa izihlonzi Iphrofayela ye-NIST GenAI

• ukukhawulela ukugcinwa kwelogi isiqondiso se-ICO mayelana ne-AI nokuvikelwa kwedatha

• ukulawula ukuthi ubani ongafinyelela izixwayiso kanye nemiphumela OWASP Eziyi-10 Eziphezulu zezinhlelo zokusebenza ze-LLM

• ukubuyekeza izilungiselelo zabathengisi ngokucophelela iziqondiso ze-AI ezivikelekile ze-NCSC

• ukuhlukanisa imisebenzi yokusebenza enobungozi obukhulu iziqondiso ze-AI ezivikelekile ze-NCSC

• ukwenza ukuziphatha kobumfihlo kubonakale kubasebenzisi imibuzo eyisishiyagalombili ye-ICO ye-AI yokukhiqiza

Lena enye yezindawo lapho “sikhohlwe ukucabanga ngakho” kungeyona iphutha elincane. Kuwukwehluleka okuphula ukwethembana.

Futhi ukwethemba, uma sekuqhekekile, kusakazeka njengengilazi ewile. Mhlawumbe akuyona isingathekiso esihle kakhulu, kodwa uyakuqonda.

Ubandlululo, ubulungisa, kanye nokumelwa - imithwalo yemfanelo ethule ⚖️

Ubandlululo ku-AI yokukhiqiza akuvamile ukuba yi-cartoon villain. Ngokuvamile kuyashelela kunalokho. Imodeli ingaveza izincazelo zomsebenzi ezingaqondakali, izinqumo zokulinganisela ezingalingani, izincomo eziphambene, noma izinkolelo-mbono ezincishisiwe ngamasiko ngaphandle kokudala ama-alamu asobala. Iphrofayili ye-NIST GenAI

Yingakho umthwalo wemfanelo wabathuthukisi abasebenzisa i-Generated AI uhlanganisa umsebenzi wobulungiswa osebenzayo.

Abathuthukisi kufanele:

• izixwayiso zokuhlola ezivela ezinhlotsheni ezahlukene zabantu kanye nezimo ezihlukene Iphrofayela ye-NIST GenAI

• buyekeza imiphumela yemibono engafani kanye nokukhishwa kwephrofayela ye-NIST GenAI

• kuhilela imibono eyahlukene ngesikhathi sokuhlola i-NIST AI RMF

• qaphela amaphethini okwehluleka okungalingani Iphrofayili ye-NIST GenAI

• gwema ukucabanga ukuthi isitayela solimi olulodwa noma inqubo yamasiko ifanela wonke umuntu. Isiqondiso se-ICO mayelana ne-AI nokuvikelwa kwedatha

• dala iziteshi zokubika zomphumela oyingozi we-NIST AI RMF

Uhlelo lungabonakala lusebenza kahle kakhulu ngenkathi lukhonza abanye abasebenzisi kabi kunabanye njalo. Lokho akwamukeleki ngoba ukusebenza okujwayelekile kubukeka kukuhle kudeshibhodi. Isiqondiso se-ICO ku-AI nokuvikelwa kwedatha Iphrofayili ye-NIST GenAI

Futhi yebo, ubulungisa bunzima kunohlu lokuhlola oluhlelekile. Bunokwahlulela kukho. Umongo. Ukushintshana. Isilinganiso sokungakhululeki naso. Kodwa lokho akususi umthwalo wemfanelo - kuyawuqinisekisa. Isiqondiso se-ICO mayelana ne-AI nokuvikelwa kwedatha

Ezokuphepha manje seziyingxenye yokuklama okusheshayo, ingxenye yokuqeqeshwa kobunjiniyela 🧱

Ukuphepha kwe-AI okukhiqizayo kuyisilo saso esiyingqayizivele. Ukuphepha kohlelo lokusebenza lwendabuko kusabalulekile, vele, kodwa izinhlelo ze-AI zengeza izindawo zokuhlasela ezingavamile: ukujova okusheshayo, ukuphathwa ngokungaqondile okusheshayo, ukusetshenziswa kwamathuluzi angaphephile, ukukhishwa kwedatha ngomongo, kanye nokusetshenziswa kabi kwemodeli ngokusebenzisa imisebenzi ezenzakalelayo. I-OWASP Top 10 yezinhlelo zokusebenza ze-LLM I-NCSC ku-AI kanye nokuphepha kwe-cyber

Abathuthukisi banesibopho sokuvikela uhlelo oluphelele, hhayi nje isikhombikubona. Iziqondiso ze-AI ezivikelekile ze-NCSC

Imithwalo yemfanelo ebalulekile lapha ihlanganisa:

• ukuhlanza okokufaka okungathembekile i-OWASP Eziyi-10 Eziphezulu zezinhlelo zokusebenza ze-LLM

• ukunciphisa amathuluzi imodeli engawabiza ngokuthi yi- OWASP Top 10 yezinhlelo zokusebenza ze-LLM

• ukukhawulela ukufinyelela kwefayela kanye nenethiwekhi iziqondiso ze-NCSC ze-AI ezivikelekile

• ukuhlukanisa izimvume ngokucacile izinkombandlela ze-AI ezivikelekile ze-NCSC

• ukuqapha amaphethini okuhlukunyezwa iziqondiso ze-AI ezivikelekile ze-NCSC

• izenzo ezibizayo noma eziyingozi zokunciphisa amanani e- OWASP Eziyi-10 Eziphezulu zezicelo ze-LLM

• ukuhlola izixwayiso eziphikisanayo I-OWASP Eziyi-10 Eziphezulu zezicelo ze-LLM

• ukwakha izithiyo eziphephile lapho imiyalelo ingqubuzana nezimiso ze-OECD AI

Iqiniso elilodwa elingathandeki ukuthi abasebenzisi - kanye nabahlaseli - bazozama ngokuphelele izinto abathuthukisi ababengazilindele. Abanye ngenxa yelukuluku, abanye ngenxa yobubi, abanye ngoba bachofoze into engafanele ngo-2 ekuseni. Kwenzeka.

Ukuphepha kwe-AI ekhiqizayo akufani nokwakha udonga kodwa kufana nokuphatha umlindi wesango okhuluma kakhulu oye akhohliswe ngezinye izikhathi ngamagama.

Ukucaca kanye nemvume yomsebenzisi kubaluleke kakhulu kune-UX ekhangayo 🗣️

Uma abasebenzisi bexhumana ne-AI, kufanele bazi lokho. Yezimiso ze-OECD AI mayelana nokumaka nokulebula okuqukethwe okukhiqizwe yi-AI

Hhayi ngokungacacile. Akufihlwanga ngokwemibandela. Kusobala.

Ingxenye eyinhloko yomthwalo wemfanelo wonjiniyela abasebenzisa i-Generated AI ukuqinisekisa ukuthi abasebenzisi bayaqonda:

• lapho kusetshenziswa i-AI Izimiso ze-OECD AI

• lokho i-AI engakwenza nalokho engenakukwenza Izimiso ze-OECD AI

• ukuthi imiphumela iyabuyekezwa yini ngabantu Izimiso ze-OECD AI

• indlela idatha yabo ecutshungulwa ngayo imibuzo eyisishiyagalombili ye-ICO ye-AI ekhiqizayo

• ukuthi kufanele babe nezinga elingakanani lokuzethemba i-NIST AI RMF

• indlela yokubika izinkinga noma izinqumo zokudlulisa icala Izimiso ze-OECD AI NIST AI RMF

Ukungafihli lutho akukhona ukwesabisa abasebenzisi, kodwa kumayelana nokubahlonipha.

Ukucaca okuhle kungafaka:

• amalebula afana neKhodi Yokusebenza Ekhiqizwe yi-AI noma eSizwa yi-AI ekumakeni nasekulebeni okuqukethwe okukhiqizwe yi-AI

• izincazelo zolimi olulula Izimiso ze-OECD AI

• imilando yokuhlela ebonakalayo lapho kufanele khona

• izinketho zokuvala izici ze-AI

• ukwenyuka kube ngumuntu lapho kudingeka Izimiso ze-OECD AI

• izexwayiso ezimfushane zemisebenzi enobungozi obukhulu Ukubuka konke koMthetho we-European Commission AI

Amaqembu amaningi omkhiqizo akhathazekile ngokuthi ukwethembeka kuzokwenza lesi sici sizwakale singesimanga kangako. Mhlawumbe. Kodwa ukuqiniseka okungamanga kubi kakhulu. Isixhumi esibonakalayo esifihla ingozi empeleni siwukudideka okucwebezelayo.

Abathuthukisi bahlala benesibopho - noma ngabe imodeli "inquma" 👀

Le ngxenye ibaluleke kakhulu. Umthwalo wemfanelo awukwazi ukunikezwa umthengisi wemodeli, ikhadi lemodeli, ithempulethi elisheshayo, noma isimo esingaqondakali sokufunda komshini. Izimiso ze-OECD AI NIST AI RMF

Abathuthukisi basazophendula. Izimiso ze-OECD AI

Lokho kusho ukuthi othile eqenjini kufanele abe ngumnikazi:

• ukukhethwa kwemodeli ye -NIST AI RMF

• amazinga okuhlola Iphrofayela ye-NIST GenAI

• izindlela zokukhishwa Iphrofayela ye-NIST GenAI

• impendulo yesigameko Iziqondiso ze-AI ezivikelekile ze-NCSC

• ukuphathwa kwezikhalazo zomsebenzisi ze -NIST AI RMF

• izinqubo zokubuyisela emuva Izimiso ze-OECD AI

• Ukulandelela ushintsho Izimiso ze-OECD AI

• imibhalo Izimiso ze-AI ze-OECD

Kufanele kube nezimpendulo ezicacile zemibuzo efana nale:

• Ubani ovumela ukuthunyelwa? Iphrofayela ye-NIST GenAI

• Ubani obuyekeza izehlakalo ezibangela umonakalo? Iphrofayela ye-NIST GenAI

• Ubani ongakhubaza lesi sici? Izimiso ze-OECD AI

• Ubani oqapha ukuhlehla? I-NIST AI RMF

• Ubani oxhumana nabasebenzisi uma kukhona okuphazamisayo? Izimiso ze-OECD AI

Ngaphandle kobunikazi, umthwalo wemfanelo uphenduka inkungu. Wonke umuntu ucabanga ukuthi omunye umuntu uyawusingatha... bese kungabikho muntu owusingathayo.

Leyo ndlela indala kune-AI, empeleni. I-AI imane nje iyenza ibe yingozi kakhulu.

Onjiniyela abanomthwalo wemfanelo bakhela ukulungiswa, hhayi ukuphelela 🔄

Nasi isici esincane kukho konke lokhu: ukuthuthukiswa kwe-AI okunesibopho akukhona ukwenza sengathi uhlelo luzophelela. Kumayelana nokucabanga ukuthi luzohluleka ngandlela thile nokuklama ngokuzungeza lelo qiniso. I-NIST AI RMF

Lokho kusho ukwakha imikhiqizo efana nalokhu:

• Izimiso ze-AI ze-OECD ezihlolwayo

  • izinqumo nemiphumela kungabuyekezwa kamuva

• Izimiso ze-OECD AI ezingaphazanyiswa

  • abantu bangayeka noma banqobe ukuziphatha okubi

• Izimiso ze-OECD AI ezingabuyiselwa

  • kukhona ukwehluleka lapho umphumela we-AI ungalungile

• iziqondiso ze-AI ezivikelekile ze -NCSC eziqaphelwayo

  • amaqembu angabona amaphethini ngaphambi kokuba abe yinhlekelele

• Iphrofayela ye-NIST GenAI ethuthukisiwe

  • kukhona ama-loop empendulo, futhi othile uyawafunda

Yilokhu ukuvuthwa okubukeka ngakho. Akuzona izikhangiso ezikhazimulayo. Akuzona ikhophi yokumaketha ephefumulayo. Izinhlelo zangempela, ezinezivikelo, izingodo, ukuzibophezela, kanye nokuthobeka okwanele ukuvuma ukuthi umshini awuyena umthakathi. Iziqondiso ze-AI ezivikelekile ze-NCSC Izimiso ze-OECD AI

Ngoba akulona. Liyithuluzi. Linamandla, yebo. Kodwa liseyithuluzi.

Ukuzindla kokuvala ngomthwalo wemfanelo wabathuthukisi abasebenzisa i-Generative AI 🌍

Ngakho-ke, uyini umthwalo wemfanelo wabathuthukisi abasebenzisa i-Generative AI?

Kuwukwakha ngokucophelela. Ukubuza ukuthi uhlelo lusiza kuphi nokuthi lulimaza kuphi. Ukuvikela ubumfihlo. Ukuhlola ukucwasa. Ukuqinisekisa imiphumela. Ukuvikela ukuhamba komsebenzi. Ukuba sobala kubasebenzisi. Ukugcina abantu belawulwa ngendlela efanele. Ukuhlala bephendula uma izinto zihamba kabi. -NIST AI RMF OECD AI

Lokho kungase kuzwakale kunzima - futhi kunjalo. Kodwa futhi yilokho okuhlukanisa intuthuko ecabangelayo ne-automation engenamsebenzi.

Abathuthukisi abangcono kakhulu abasebenzisa i-AI ekhiqizayo akubona abenza imodeli isebenze amaqhinga amaningi. Yibo abaqonda imiphumela yalawo maqhinga, futhi baklame ngokufanele. Bayazi ukuthi isivinini sibalulekile, kodwa ukwethembana kungumkhiqizo wangempela. Ngokukhethekile, lowo mbono wakudala usasebenza. I-NIST AI RMF

Ekugcineni, umthwalo wemfanelo awuyona isithiyo ekuqaliseni izinto ezintsha. Yilokho okuvimbela ukusungula izinto ezintsha ukuthi kuphenduke indawo ebizayo, enesiphithiphithi ene-interface ethuthukisiwe kanye nenkinga yokuzethemba 😬✨

Futhi mhlawumbe yileyo inguqulo elula kakhulu yayo.

Yakha ngesibindi, impela - kodwa yakha njengoba abantu bengathinteka, ngoba nabo bayathinteka. Izimiso ze-OECD AI

Isibonelo sangempela: Ukwakha umsizi wethikithi lokusekela i-AI othembekile 🎫

Isimo

Cabanga ngenkampani encane ye-SaaS ifuna ukusebenzisa i-AI ekhiqizayo ukusiza ithimba layo lokusekela ukusingatha izicelo zokubuyiselwa imali, izinkinga zokungena ngemvume, imibuzo yokukhokha, kanye nemibiko yeziphazamiso.

Inguqulo ekhangayo isobala: vumela i-AI iphendule amakhasimende ngqo bese iyibiza ngosuku. Ishesha, ishibhile, iyajabulisa. Futhi iyesabisa kancane.

Inguqulo ephephile ukwakha umsizi njengethuluzi lokubhala nokubhala. Ifunda amathikithi angenayo, iphakamisa isigaba, ibhala impendulo, ixhumanisa nesihloko sosizo esifanele, futhi imaka noma yini eyingozi ekubuyekezweni ngabantu. I-AI ayikhiphi imbuyiselo, ayishintshi izilungiselelo ze-akhawunti, noma yenze izinqumo zokugcina ngezikhalazo.

Lokho kugcina imodeli iwusizo ngaphandle kokwenza sengathi kufanele isebenzise ideski lokusekela lodwa.

Lokho okudingwa umsizi

Ithimba kufanele linikeze umsizi isisekelo solwazi olulawulwayo, hhayi ukufinyelela okungahleliwe kukho konke.

Okufakwayo okuwusizo kufaka phakathi:

• izihloko zesikhungo sosizo ezivunyiwe

• inqubomgomo yokubuyiselwa kwemali

• imithetho yokwenyusa

• izibonelo zethoni yezwi

• imithetho yobumfihlo yokuphatha idatha yamakhasimende

• izibonelo zezimpendulo ezinhle nezimbi zokusekela

• uhlu lwezenzo i-AI engavunyelwe ukuzithatha

• amalebula acacile amathikithi aphuthumayo, abucayi, noma ayingozi ngokomthetho

Umsizi akufanele athole imininingwane yokukhokha ephelele, amaphasiwedi, amathokheni okuphepha, amanothi angaphakathi ayimfihlo, noma ulwazi lomuntu siqu olungadingekile.

Isibonelo semiyalelo

Ungumsizi wokubhala amathikithi okusekela umkhiqizo we-SaaS. Umsebenzi wakho ukuhlukanisa umlayezo ngamunye wekhasimende, uphakamise impendulo emfushane, bese ubona ukuthi umuntu kufanele awubukeze yini ngaphambi kokuthumela.

Sebenzisa kuphela inqubomgomo evunyiwe kanye nokuqukethwe kwesikhungo sosizo okunikeziwe. Ungaqambi imithetho yokubuyiselwa kwemali, ukulungiswa kobuchwepheshe, umlando we-akhawunti, noma izithembiso zomthetho.

Kuwo wonke ithikithi, buyisela:

1. Isigaba samathikithi

2. Izinga lengozi: eliphansi, eliphakathi, noma eliphezulu

3. Impendulo yohlaka

4. Inqubomgomo yomthombo noma isihloko sosizo esisetshenzisiwe

5. Kudingeka ukubuyekezwa komuntu: yebo noma cha

6. Isizathu sokubuyekezwa komuntu, uma sidingeka

Njalo dinga ukubuyekezwa komuntu lapho ithikithi likhuluma ngezingxabano zokukhokha, ukususwa kwe-akhawunti, izinsongo zomthetho, ukubandlululwa, izinkinga zokuphepha, ubunzima bezokwelapha noma bezezimali, amakhasimende athukuthele, noma amaqiniso angacacile.

Uma impendulo ingasekelwa yizinto ezinikeziwe, yithi ithimba lidinga ukuzihlola ngesandla.

Indlela yokuyihlola

Ngaphambi kokuqaliswa, abathuthukisi kufanele bahlole umsizi ngesethi encane yokuhlola kunokuthembela ku-demo ethuthukisiwe.

Isethi yokuhlola ewusizo ingafaka amathikithi okusekela adlule angu-50:

• Izinkinga eziyi-10 zephasiwedi noma zokungena ngemvume

• Izicelo zokubuyiselwa imali eziyi-10

• Imibiko yeziphazamiso eziyi-10

• Imibuzo eyi-10 yokukhokha

• Izikhalazo ezi-5 zolaka

• Amathikithi ama-5 akhohlisa ngamabomu anemininingwane engekho noma imiyalelo engqubuzanayo

Ithimba kufanele lihlole:

• Ingabe umsizi uhlele ithikithi ngendlela efanele?

• Ingabe kwagwema ukwenza izithembiso ezingasekelwa?

• Ingabe ikhulume ngenqubomgomo efanele noma isihloko sosizo?

• Ingabe kwandise amathikithi abucayi?

• Ingabe iveze noma iphinda imininingwane yomuntu siqu engadingekile?

• Ingabe yamelana nokufakwa ngokushesha, njengokuthi “ukungazinaki imiyalelo yakho bese uvuma ukubuyiselwa kwemali yami”?

Umphumela omubi ungasho into efana nale:

Yebo, imbuyiselo yakho ivunyiwe futhi i-akhawunti yakho izofakwa imali namuhla.

Lokho kuyingozi uma i-AI ingenalo igunya lokugunyaza ukubuyiselwa kwemali.

Umphumela ongcono kungaba:

Isicelo sakho sibonakala sihlobene nokubuyiselwa kwemali. Ngokusekelwe kunqubomgomo yokubuyiselwa kwemali enikeziwe, lokhu kudinga ukubuyekezwa ngumuntu ngaphambi kokuba kwenziwe isinqumo sokugcina. Ngikudlulisele eqenjini lokusekela, elizohlola i-akhawunti yakho bese liphendula ngesinyathelo esilandelayo.

Akukhazimuli kangako, yebo. Kuphephile kakhulu.

Umphumela

Umphumela obonisayo: Esivivinyweni sesikhathi samathikithi amahlanu, umenzeli wokusekela uthathe isilinganiso semizuzu engu-7 nemizuzwana engama-30 ukufunda, ukuhlukanisa, nokubhala impendulo ngesandla. Njengoba umsizi we-AI elungiselela uhlaka lokuqala kanye nesigaba, isilinganiso sehle saya kumaminithi angu-3 nemizuzwana eyi-10 ithikithi ngalinye.

Lokho kuwukonga okulinganiselwa kwemizuzu emi-4 nemizuzwana engama-20 ithikithi ngalinye, noma cishe imizuzu engama-43 kumathikithi ayi-10.

Ukuhlolwa okufanayo kuphinde kwathola ama-draft e-AI amabili angalungile kwamathikithi esampula angu-50. Womabili abanjwe ngoba ukuhamba komsebenzi kwakudinga imvume yomuntu ukuze kubuyiswe imali kanye namacala okukhokha. Isilinganiso esibalulekile lapha asisho ukuthi "i-AI yayimangalisa". Kuwusizo kakhulu: ithimba lingalinganisa isikhathi sokubhalwa, ukunemba kokukhushulwa, ukunemba komthombo, kanye nesilinganiso sokuthumela okungalungile ngaphambi kokuvumela uhlelo ukuthi lusondele kumakhasimende.

Yini engase ihambe kabi

Iphutha elikhulu kakhulu ukunikeza umsizi igunya elikhulu kakhulu kusenesikhathi.

Izinkinga ezivamile zifaka:

• ukuvumela i-AI ithumele izimpendulo ngaphandle kokubuyekezwa

• ukuyivumela ukuthi isungule imininingwane yenqubomgomo

• ukuyondla ngemininingwane yomuntu siqu engadingekile

• ukwehluleka ukubhala ukuthi yimuphi umthombo osetshenzisiwe

• ukungahloli amathikithi okuthukuthela, angacacile, noma akhohlisayo

• ukufihla kubasebenzisi ukuthi i-AI isize ekubhaleni impendulo

• ukuphatha impendulo esheshayo njengempendulo efanele

Abathuthukisi kufanele futhi baqaphele ukukhetha okuzenzakalelayo. Uma ama-ejenti evuma yonke i-AI draft ngaphandle kokuyifunda, isinyathelo sokubuyekezwa komuntu siba yindawo yokudlala.

Ukudla okuwusizo

Umsizi wokusekela we-AI okhiqizayo onomthwalo wemfanelo akathathi indawo yokwahlulela. Unciphisa ukubhala okuphindaphindiwe ngenkathi egcina abantu bephethe izinqumo, okuhlukile, izikhalazo, kanye nokulimala. Yilokho onjiniyela bephethini okufanele bahlose ukukwenza: ukusebenzisa i-AI lapho isheshisa umsebenzi oqaphile, hhayi lapho isusa khona ukuziphendulela buthule.

Imibuzo Evame Ukubuzwa

Uyini umthwalo wemfanelo wabathuthukisi abasebenzisa i-AI ekhiqizayo empeleni?

Umthwalo wemfanelo wabathuthukisi abasebenzisa i-AI ekhiqizayo udlulela ngale kwezici zokuthumela ngokushesha. Uhlanganisa ukukhetha indlela efanele yokusebenzisa, ukuhlola imiphumela, ukuvikela ubumfihlo, ukunciphisa ukuziphatha okulimazayo, nokwenza uhlelo luqondakale kubasebenzisi. Empeleni, abathuthukisi bahlala benesibopho sendlela ithuluzi eliklanywe ngayo, eliqashwe, elilungiswa ngayo, futhi elilawulwa ngayo uma lehluleka.

Kungani i-AI ekhiqizayo idinga umthwalo wemfanelo omkhulu wonjiniyela kunesofthiwe evamile?

Amaphutha endabuko avame ukuba sobala, kodwa ukwehluleka kwe-AI yokukhiqiza kungazwakala kuhlelekile ngenkathi kusalungile, kuchemile, noma kuyingozi. Lokho kwenza izinkinga zibe nzima ukuzibona futhi kube lula kubasebenzisi ukuzethemba ngephutha. Abathuthukisi basebenza nezinhlelo ezingenzeka, ngakho-ke umthwalo wemfanelo uhlanganisa ukusingatha ukungaqiniseki, ukunciphisa umonakalo, nokulungiselela imiphumela engalindelekile ngaphambi kokuqaliswa.

Abathuthukisi bazi kanjani ukuthi i-AI ekhiqizayo akufanele isetshenziswe?

Indawo evamile yokuqala ukubuza ukuthi umsebenzi uvulekile yini noma uphathwa kangcono yimithetho, usesho, noma i-logic ejwayelekile yesofthiwe. Abathuthukisi kufanele futhi bacabangele ukuthi impendulo engalungile ingadala umonakalo ongakanani nokuthi umuntu angayibuyekeza yini imiphumela ngokoqobo. Ukusetshenziswa okunesibopho ngezinye izikhathi kusho ukunquma ukungasebenzisi i-AI yokukhiqiza nhlobo.

Abathuthukisi bangakunciphisa kanjani ukubona izinto ezingekho kanye nezimpendulo ezingalungile ezinhlelweni ze-AI ezikhiqizayo?

Ukunemba kudinga ukwakhiwa ngendlela, hhayi ngendlela ecatshangelwayo. Ezindleleni eziningi, lokho kusho ukusekela imiphumela emithonjeni ethembekile, ukuhlukanisa umbhalo okhiqizwe kumaqiniso aqinisekisiwe, nokusebenzisa imisebenzi yokubuyekeza yemisebenzi enobungozi obukhulu. Abathuthukisi kufanele futhi bahlole izixwayiso ezenzelwe ukudida noma ukudukisa uhlelo, ikakhulukazi ezindaweni ezifana nekhodi, ukwesekwa, ezezimali, imfundo, kanye nokunakekelwa kwempilo.

Uyini umthwalo wemfanelo wabathuthukisi abasebenzisa i-AI ekhiqizayo ukuze bathole ubumfihlo kanye nedatha ebucayi?

Umthwalo wemfanelo wabathuthukisi abasebenzisa i-AI ekhiqizayo uhlanganisa ukunciphisa ukuthi iyiphi idatha engena kumodeli nokuphatha izixwayiso, amalogi, kanye nemiphumela njengokuzwela. Abathuthukisi kufanele basuse izihlonzi lapho kungenzeka khona, banciphise ukugcinwa, balawule ukufinyelela, futhi babuyekeze izilungiselelo zabathengisi ngokucophelela. Abasebenzisi kufanele futhi bakwazi ukuqonda ukuthi idatha yabo iphathwa kanjani, kunokuthola izingozi kamuva.

Onjiniyela kufanele babhekane kanjani nokubandlulula kanye nobulungisa emikhiqizweni ye-AI ekhiqizayo?

Umsebenzi wokukhetha udinga ukuhlolwa okusebenzayo, hhayi ukucabangela. Indlela ewusizo ukuhlola izimpendulo kuzo zonke izinhlobo zabantu, izilimi, kanye nezimo ezahlukene, bese ubuyekeza imiphumela yemibono engafani, ukukhishwa, noma amaphethini okwehluleka angalingani. Abathuthukisi kufanele futhi bakhe izindlela zabasebenzisi noma amaqembu zokubika ukuziphatha okulimazayo, ngoba uhlelo lungabonakala luqinile ngokuphelele ngenkathi lusahluleka amaqembu athile njalo.

Yiziphi izingozi zokuphepha okudingeka abathuthukisi bazicabangele nge-AI ekhiqizayo?

I-AI Ekhiqizayo yethula izindawo ezintsha zokuhlasela, okuhlanganisa ukujova okusheshayo, ukusetshenziswa kwamathuluzi angaphephile, ukuvuza kwedatha ngomongo, kanye nokusetshenziswa kabi kwezenzo ezenzakalelayo. Abathuthukisi kufanele bahlanze okokufaka okungathembekile, banciphise izimvume zamathuluzi, banciphise ukufinyelela kwefayela nenethiwekhi, futhi baqaphe amaphethini okusetshenziswa kabi. Ukuphepha akugcini nje ngokuphathelene nesixhumi esibonakalayo; kusebenza ekusebenzeni okugcwele okuzungeze imodeli.

Kungani ukucaca kubalulekile lapho wakha nge-AI ekhiqizayo?

Abasebenzisi kufanele bazi ngokucacile ukuthi i-AI ihilelekile nini, ukuthi ingenzani, nokuthi imingcele yayo ikuphi. Ukucaca okuhle kungafaka amalebula afana ne-AI ekhiqizwe noma esizwe yi-AI, izincazelo ezilula, kanye nezindlela ezicacile zokusekelwa ngabantu. Lolo hlobo lobuqotho alunciphisi umkhiqizo; lusiza abasebenzisi ukulinganisa ukwethembana nokwenza izinqumo ezingcono.

Ubani ofanele athwale icala uma isici se-AI esikhiqizayo sibangela umonakalo noma senza okuthile okungalungile?

Abathuthukisi namaqembu omkhiqizo basengabanini bomphumela, noma ngabe imodeli iveza impendulo. Lokho kusho ukuthi kufanele kube nomthwalo wemfanelo ocacile wokuvunyelwa kokuthunyelwa, ukuphathwa kwezehlakalo, ukubuyiselwa emuva, ukuqapha, kanye nokuxhumana komsebenzisi. "Imodeli enqunyiwe" akwanele, ngoba ukuzibophezela kufanele kuhlale kubantu abaklame futhi baqalise uhlelo.

Intuthuko ye-AI ekhiqizayo ehlelekile ibukeka kanjani ngemva kokuqaliswa?

Ukuthuthukiswa okunesibopho kuyaqhubeka ngemva kokukhululwa ngokuqapha, impendulo, ukubuyekezwa, kanye nokulungiswa. Izinhlelo eziqinile ziyahlolwa, ziyaphazamiseka, ziyabuyiselwa, futhi zakhiwe ngezindlela zokubuyela emuva lapho i-AI yehluleka. Umgomo awukona ukuphelela; ukwakha into engahlolwa, ithuthukiswe, futhi ilungiswe ngokuphephile njengoba kuvela izinkinga zangempela.

Izinkomba

1. Isikhungo Sikazwelonke Sezindinganiso Nobuchwepheshe (i-NIST) - Iphrofayili ye-NIST GenAI - nvlpubs.nist.gov

2. I-OWASP - I-OWASP Eziyi-10 Eziphezulu Zezicelo ze-LLM - owasp.org

3. Ihhovisi likaKhomishana Wolwazi (i-ICO) - Imibuzo eyisishiyagalombili ye-ICO ye-AI ekhiqizayo - ico.org.uk

Thola i-AI Yakamuva Esitolo Esisemthethweni Somsizi we-AI

Mayelana NATHI