imvamisa iqala ngendlela efanayo. Emhlanganweni wobuholi, othile uthi: “Ake sisebenzise i-AI!” Amakhanda anikine, umdlandla akha, nangaphambi kokuba wazi, amakamelo afinyelela esiphethweni esizenzakalelayo: “Impela – sizokwakha i-chatbot.” Lowo mqondo uyaqondakala. Amamodeli amakhulu olimi anamandla, abuquitous, futhi ahehayo. Bathembisa ukufinyelela okunembile kulwazi lwendawo yonke nokusebenza.
Iqembu lihamba liqala ukwakha. Ngokushesha, kuza isikhathi sedemo. Kuvela isikhombimsebenzisi sokuxoxa esibunjiwe, sihambisana nezimpikiswano zokuzethemba mayelana nokuthi kungani ngalesi sikhathi, kuzohluka. Ngaleso sikhathi, noma kunjalo, imvamisa ayifinyelelanga kubasebenzisi bangempela ezimweni zangempela, futhi ukuhlola kuhlotshiswe futhi kunethemba. Othile ezilalelini ngokungenakuqhathaniswa ukhuphuka nombuzo ngokwezifiso, ukucasula ama-bot. Abathuthukisi bathembisa ukukulungisa “it”, kepha ezimweni eziningi, inkinga eyisisekelo is systemic.
Lapho i-Chatbot ishaya phansi, ithemba lokuqala livame ukufana ngokukhungatheka komsebenzisi. Lapha, izinto zithola umuntu uqobo ngoba emasontweni adlule, ngaphoqeleka ukuba ngichithe isikhathi sikhuluma nezingxoxo ezihlukile. Ngijwayele ukubambezela ukusebenzisana nabanikezeli bezinsizakalo kuze kube yilapho isimo sibe okungagcineki, futhi amacala ambalwa awohlokile. Amawijethi we-Chatbot amamatheka waba yithemba lami lokugcina ngaphambi kwekholi yaphakade yocingo, kepha:
- Ngemuva kokungena kwisiza se-insurer yemoto yami, ngacela ukuchaza ukwanda kwentengo okungachazwanga, kuphela ukubona ukuthi i-chatbot ayinakho ukufinyelela kwimininingwane yami yamanani. Konke okungakunikeza kwakuyinombolo yocingo. Ouch.
- Ngemuva kokuthi indiza ikhanseliwe ngomzuzu wokugcina, ngabuza i-chatbot yendiza ngesizathu. Kuxolise ngenhlonipho ukuthi, ngoba isikhathi sokuhamba sasisedlule, sasingangisiza. Kwakuvulekele ukuxoxa ngazo zonke ezinye izihloko.
- Esizeni se-telco, ngabuza ukuthi kungani uhlelo lwami lweselula seluphelelwe yisikhathi ngokuzumayo. I-Chatbot yaphendula ngokuzethemba ukuthi ayikwazi ukuphawula ngezinto zesivumelwano futhi yangidlulisela kuma-FAQs. Njengoba bekulindelekile, lokhu bekuyinto ende kodwa akunangqondo.
Lokhu kuhlangana akuzange kusondeze ikhambi kwangishiya ngakunye okuphambene nenjabulo. Izingxoxo zazizwa njengezimbali zakwamanye amazwe. Behleli lapho, badla indawo ethengiswayo, i-latency, nokunakwa, kepha azange bangeze inani.
Mase weqe le mpikiswano yokuthi lawa amaphethini amnyama anhloso. Iqiniso ngukuthi, amathoyizi wezinhlelo zefa njengale engenhla ethwala umthwalo osindayo wokungenayo. Beza namathani wedatha eyingqayizivele, ulwazi, kanye nomongo. Okwamanje lapho uzama ukubahlanganisa nabo nge-LLM enenjongo ejwayelekile, wenza imihlaba emibili ixabana. Imodeli idinga ukungenisa umongo womkhiqizo wakho ukuze icabange ngokunenjongo ngesizinda sakho. Ubunjiniyela obufanele bomongo budinga ikhono nesikhathi sokuhlola okungapheli kanye ne-iteration. Futhi ngaphambi kokuthi ufike kulelo phuzu, idatha yakho idinga ukulungele, kepha ezinhlanganweni eziningi, idatha inomsindo, ihlukaniswe, noma ilahlekile nje.
Kulokhu okuthunyelwe, ngizophinda ngithole ukuqonda kusuka encwadini yami ubuciko bokuthuthukiswa komkhiqizo we-AI kanye nenkulumo yami yakamuva engqungqutheleni ye-Google AI ye-AI futhi wabelane nge-AI emikhiqizweni ekhona.
Kusetshenziswa amamodeli amancane wengozi ephansi, ukuhlanganiswa kwe-AI okukhulayo
“Lapho usebenzisa i-AI, ngibona izinhlangano eziningi zihluleka ngokuqala okukhulu kakhulu kunokuqala okuncane kakhulu.” ( Andrew ng).
Isikhathi sokuhlanganisa i-AI:
- Iqembu lakho lobuchwepheshe lidinga ukulungiselela idatha futhi lifunde amasu atholakalayo namathuluzi.
- Udinga ukuProtatype futhi u-Interate ukuthola izindawo ezimnandi zenani le-AI kumkhiqizo wakho nasemakethe yakho.
- Abasebenzisi badinga ukulinganisa ukwethenjwa kwabo lapho beqhubekela phambili kokuhlangenwe nakho okusha kwe-probabike.
Ukuzivumelanisa nalezi zindophi zokufunda, akufanele usheshe ukudalula i-AI – ikakhulukazi ukusebenza kwengxoxo evulekile – kubasebenzisi bakho. I-AI yethula ukungaqiniseki namaphutha kusipiliyoni, iningi labantu alithandi.
Indlela eyodwa ephumelelayo yokupheka uhambo lwakho lwe-AI kumongo we-brownfield ngokusebenzisa amamodeli amancane olimi (i-SLMS), okuhlukahluka kusuka ezigidini ezingamakhulu ambalwa kuya kwamapharamitha ayizigidi eziyizinkulungwane ezimbalwa. Bangakwazi ukuhlangana kalula ngemininingwane ekhona yomkhiqizo kanye nengqalasizinda, kunokuba ungeze ngaphezulu kwezobuchwepheshe.
Kuqeqeshwa kanjani ama-SLMS
Iningi le-SLMS litholakala kumamodeli amakhulu ngokusebenzisa i-distillation yolwazi. Kulokhu kusethwa, imodeli enkulu isebenza njengothisha futhi omncane njengomfundi. Isibonelo, i-Gemini's Gemini isebenze njengothisha weGemma 2 noGemma 3, kanti uLlama Behemoth waqeqesha umhlambi wawo wama-LLama 4 amamodeli amancane. Njengoba nje uthisha ongumuntu othanda ukufundela izincazelo ezicacile nezifundo ezihlelekile, imodeli enkulu iyekela isikhala sayo esikhulu sepharamitha ibe yindawo encane, eyethulwe ngumfundi engayithola. Umphumela uyimodeli ecwebezelayo egcina amakhono othisha kepha asebenza ngamapharamitha ambalwa kakhulu futhi izindleko eziphansi kakhulu zokuncintisana.
Usebenzisa i-SLMS
Enye yezinzuzo ezisemqoka ze-SLMS ukuguquguquka kwabo kokuthunyelwa. Ngokungafani ne-LLMS esetshenziswa kakhulu nge-API yangaphandle, amamodeli amancane angaqhutshwa endaweni yakini, kungaba kwingqalasizinda yenhlangano yakho noma ngqo kudivayisi yomsebenzisi:
- Ukuhanjiswa Kwendawo: Ungasingatha i-SLMS kumaseva akho noma ngaphakathi kwendawo yakho yamafu, ukugcina ukulawula okugcwele kwedatha, i-latency, kanye nokuhambisana. Lokhu kusetha kulungele izinhlelo zokusebenza zebhizinisi lapho imininingwane ebucayi noma izinkinga zokulawula zenza ama-apis avela eceleni angenzi lutho.
Ukuhanjiswa Kwendawo futhi kukunikeza namathuba okuguqula okuhle okuhle Njengoba uqoqa idatha ethe xaxa futhi udinga ukuphendula ekukhuleni okulindelwe kwabasebenzisi.
- Ukuthunyelwa kwensiza ngesiphequluli: Iziphequluli zanamuhla zinamakhono e-AI angakwazi ukuncika kuwo. Isibonelo, i-Chrome ihlanganisa i-Gemini nano nge-AI APIS eyakhelwe ngaphakathi, kanti onqenqemeni lweMicrosoft luhlanganisa i-PHI-4 (bheka imibhalo esheshayo ye-API). Amamodeli wokugijima ngokuqondile esipheqululini anika amandla ama-latency aphansi, ubumfihlo – ukulondolozwa kokusetshenziswa kwamacala njengokuphakanyiswa kombhalo ohlakaniphile, ifomu autofill, noma usizo lwesimo.
Uma ungathanda ukufunda kabanzi ngobuchwepheshe be-SLMS, nazi izinsiza ezimbalwa eziwusizo:
Manje ake siqhubeke futhi sibone ukuthi yini ongayenza nge-SLMS ukuhlinzeka ngenani lomsebenzisi futhi uthuthuke ngokuqinile ekuhlanganiseni kwakho kwe-AI.
Amathuba womkhiqizo we-SLMS
I-SLMS ikhanye emisebenzini egxile, echazwe kahle lapho umongo nedatha isivele yaziwa – izinhlobo zokusebenzisa amacala ahlala ekujuleni kwemikhiqizo ekhona. Ungacabanga ngabo njengobuhlakani obukhethekile, oshumekiwe kunokuba abasizi abajwayelekile. Ake sihambe ngamabhakede aphambili amathuba abavula e-Brownfield, njengoba kuboniswe esihlahleni samathuba alandelayo.
1. Okungcono Ukuhlaziywa komkhiqizo
Ngaphambi kokudalula izici ze-AI kubasebenzisi, funa izindlela zokuthuthukisa umkhiqizo wakho kusuka ngaphakathi. Imikhiqizo eminingi esevele ikhiqize ukusakazwa okuqhubekayo kombhalo ongahleliwe – izingxoxo zokusekela, izicelo zosizo, impendulo engaphakathi nohlelo. I-SLMS ingahlaziya le datha ngesikhathi sangempela kanye nokuqonda okungaphezulu kwe-Insights eyazisa zombili izinqumo zomkhiqizo kanye nesipiliyoni somsebenzisi esiseduze. Nazi ezinye izibonelo:
- Ithegi kanye nezingxoxo zokusekelwa komzila Njengoba zenzeka, ziqondisa izingqinamba zobuchwepheshe emaqenjini afanele.
- I-Churn amasignali ngesikhathi seseshinikushukumisela ukungenelela okufika ngesikhathi.
- Phakamisa okuqukethwe noma izenzo ezifanele kususelwa kumongo wamanje womsebenzisi.
- Thola amaphuzu aphindaphindayo wokushayisana Ngenkathi umsebenzisi esesekuhambeni, hhayi amasonto kamuva ekutholeni kabusha.
Lawa manzi angaphakathi agcina ebeka engcupheni ngenkathi engeza inani nokunikeza isikhathi seqembu lakho ukuze ufunde. Baqinisa isisekelo sakho sedatha futhi bakulungiselele ukuthola izici ezibonakalayo ezibonakalayo, ezibheke e-AI ezibheke phansi komgwaqo.
2. Susa ukungqubuzana
Okulandelayo, thatha igxathu lokubuyela emuva bese ufunda isikweletu se-UX esevele sikhona. E-Brownfield, imikhiqizo eminingi ayiyona into oyiphupho lomqambi. Benzelwe ngaphansi kwezinkinga zobuchwepheshe nezokwakha zesikhathi sazo. Nge-AI, manje sinethuba lokuphakamisa ezinye zalezo zingqinamba, zinciphisa ukungezwani futhi zidale okuhlangenwe nakho okusheshayo, okunembile.
Isibonelo esihle ukuhlunga okuhlakaniphile kumawebhusayithi asuselwa ekufundeni afana ne-ook.com. Ngokwesiko, la makhasi asebenzisa uhlu olude lwamabhokisi okuhlola kanye nezigaba ezizama ukumboza konke okuthandwa ngumsebenzisi. Bayinkimbinkimbi ukuklama nokusebenzisa, nasekupheleni, abasebenzisi abaningi abakwazi ukuthola isimo esibalulekile kubo.
Ukuhlunga okususelwa olimini kushintsha lokhu. Esikhundleni sokuzulazula nge-taxonomil eyinkimbinkimbi, abasebenzisi bamane bathayiphe lokho abakufunayo (ngokwesibonelo “amahhotela anezilwane ezifuywayo eduze nasebhishi”), futhi imodeli ihumusha ngayo engcebeni ehlelekile ngemuva kwezigcawu.
Kabanzi kabanzi, funa izindawo kumkhiqizo wakho lapho abasebenzisi badinga ukusebenzisa i-logic yakho yangaphakathi – izigaba zakho, izakhiwo, noma isigama – esikhundleni sokusebenzisana kolimi lwemvelo. Noma nini lapho abasebenzisi bengakhombisa inhloso ngqo, ususa ungqimba lokungqubuzana kwengqondo futhi wenze umkhiqizo ube nobungcweti nobungani.
3. Ukukala
Ngokuhlangenwe nakho kwakho komsebenzisi kuye kwahlaselwa, yisikhathi sokucabanga ngokwenza izinto – Ukungeza amakhono amancane, awusizo we-AI kumkhiqizo wakho. Esikhundleni sokuphinda usebenzise isipiliyoni esiyisisekelo, bheka ukuthi yini abasebenzisi esevele benza umkhiqizo wakho – imisebenzi eseceleni, ama-Worparounds, noma amathuluzi angaphandle abathembela kuwo ukuze afinyelele umgomo wawo. Ingabe amamodeli we-AI agxile asiza ukuthi akwenze ngokushesha noma ngobuchule?
Isibonelo, uhlelo lokusebenza lokuhamba lungakwazi ukuhlanganisa i-Continextual Trip Pate Generator efingqa imininingwane yokuhamba noma imiyalezo ehleliwe yabahambi. Ithuluzi lokukhiqiza lingafaka i-generator ye-RecAp yeMhlangano efingqa izingxoxo noma izinto zezenzo ezivela kumanothi wombhalo, ngaphandle kokuthumela idatha efwini.
Lezi zici zikhula ngokomzimba ekuziphatheni komsebenzisi wangempela futhi zinwebe umongo womkhiqizo wakho esikhundleni sokuchaza kabusha.
4. Yenza ngezifiso
Ukwenza okuphumelelayo komuntu siqu yi-Grail engcwele ye-AI. Idonsa amandla endabuko: Esikhundleni sokucela abasebenzisi ukuthi bafunde futhi bazivumelanise nomkhiqizo wakho, umkhiqizo wakho manje uzivumelanisa nabo njengegilavu efanelekile.
Uma uqala, gcina ukuvelela eBay – awudingi umsizi oguqukayo ngokuphelele. Esikhundleni salokho, wethula izinguquko ezincane, eziphansi kakhulu kulokho abasebenzisi abakubonayo, imininingwane ivezwa kanjani, noma yiziphi izinketho ezivela kuqala. Ezingeni lokuqukethwe, i-AI ingavumelanisa ithoni nesitayela, njengokusebenzisa igama elifushane lochwepheshe kanye nokuchazwa okuningi okuchazayo kwamagama abafikayo. Ezingeni lokuhlangenwe nakho, kungadala izindawo zokusebenza eziguquguqukayo. Isibonelo, ithuluzi lokuphatha amaphrojekthi lingahlelela izenzo ezifanele kakhulu (“dala umsebenzi,” “Vuselela Ukuvuselelwa,” “Qalisa ukufingqa”) kususelwa ekuhambeni komsebenzi okwedlule komsebenzisi.
⚠️ Lapho Ukwenza Kuthandwa Kuhamba Ngezifiso, kudalusa ukuthembana. Abasebenzisi banengqondo ukuthi bathengise imininingwane yomuntu siqu yesipiliyoni esingazizwa kangcono. Ngakho-ke, wethula owenziwe ngezifiso kuphela uma idatha yakho isilungele ukuxhasa.
Kungani “amancane” awina ngokuhamba kwesikhathi
Isici ngasinye se-AI esiphumelele – kungaba ngcono kwe-Analytics, i-UX yokuhlaziya i-UX TouchPoint, noma igxathu elenzelwe wena uqobo ngokugeleza okukhudlwana – iqinisa i-Idatha Yakho Yemininingwane futhi yakha i-TEASTION's WETICEEN CUSTION NAMA-AI WOKUFUNDA. Iphinde ibe nesisekelo sezicelo eziyinkimbinkimbi, eziyinkimbinkimbi ngokuhamba kwesikhathi. Lapho izici zakho ze- “ezincane” zisebenza ngokuthembekile, ziba yizakhi ezivuselelweyo ekuhambeni kwama-gergerflows noma amasistimu amenzeli (cf. Amaphepha ama-Nvidia amamodeli amancane olimi yikusasa le-Agentic AI).
Ukufingqa:
✅ Qala okuncane – Bheka ukuthuthukiswa kancane kancane kokuphazamiseka.
✅ Isivivinyo Ngokushesha – Amamodeli amancane asho izindleko eziphansi kanye nama-loops okudla ngokushesha.
✅ Qaphela – Qala ngaphakathi; Yethula i-AI ebhekene nomsebenzisi uma usuyiqinisekisile.
✅ Yakha imisipha yakho ye-iteration – I-Steady, inqubekela phambili ehlanganisiwe ishaya amaphrojekthi ekhanda.
Ekuqaleni kwashicilelwa ku- https://jannalipenkova.substack.com.
