Ikota yokugcina yango-2025, sekuyisikhathi sokubuyela emuva futhi uhlole izitayela ezizobumba idatha ne-2026.
Ngenkathi izihloko zingagxila ekukhishweni kwemodeli yakamuva kanye nezimpi zeBenchmark, zikude kakhulu nentuthuko eguqukayo kakhulu emhlabathini. Ushintsho lwangempela ludlala emigodini – lapho ososayensi bedatha, idatha + AI Enjiniyela, namaqembu we-AI / ML asebenzisa lezinhlelo eziyinkimbinkimbi nobuchwepheshe bokukhiqizwa. Futhi ngokusobala, ukucindezela ekwakhiweni kwe-AI – kanye namakhanda alandelayo alandelayo ekuqhubekeni komkhumbi.
Nazi izindlela eziyishumi ezichaza lokhu kuziphendukela kwemvelo, nokuthi zisho ukuthini zingena kwikota yokugcina ka-2025.
1. “Idatha + AI Abaholi” bakhuphuka
Uma ngabe uke ku-LinkedIn muva nje, kungenzeka ukuthi uke wabona ukukhuphuka okusolisayo ngenani lemininingwane yedatha + AI ku-NewsFeed yakho – ngisho naphakathi kwamalungu eqembu lakho.
Cha, akubanga nokwakhiwa kabusha ongayazi ngakho.
Ngenkathi lokhu kungukushintshwa kokuzithandela phakathi kwalabo okuhlukaniswe ngokwesiko njengedatha noma Ochwepheshe be-AI / ML, lolu shintsho ezihlokweni lubonisa iqiniso emhlabathini uMonte Carlo abelokhu ekhuluma ngalo cishe unyaka manje-idatha ne-AI alisasekho iziqondiso ezimbili ezihlukile.
Kusuka kwizisetshenziswa namakhono abawadinga ezinkingeni abazixazululayo, idatha kanye ne-AI zizinhlangothi zombili zohlamvu lwemali. Futhi ukuthi iqiniso linomthelela obonakalayo ngendlela amaqembu womabili nobuchwepheshe abaye bavela ngo-2025 (njengoba maduze uzobona).
2. Ukuxoxa nge-BI kuyashisa – kepha kudinga isheke lokushisa
I-Data Democratization ibilokhu ihamba ngefomu elilodwa noma enye enye cishe ishumi leminyaka manje, futhi ukuguqulwa kwe-BI kuyisahluko sakamuva kuleyo ndaba.
Umehluko phakathi kokuguqulwa kwe-BI BO futhi wonke amanye amathuluzi e-BI ashesha futhi ubuhle lapho athembisa ukuletha kulowo mbono wase-Utopian – ngisho nabasebenzisi bezobuchwepheshe abangewona wonke.
Isisekelo silula: Uma ungayicela, ungayifinyelela. Kuyindlela yokuwina yabanikazi nabasebenzisi ngokufanayo … ngombono. Inselelo (njengayo yonke imizamo yentando yeningi) ayilona ithuluzi ngokwalo – ukuthembeka kwento oyithandayo.
Ukuphela kwento embi kakhulu kunokuqonda okubi okuwukuqonda okubi okulethwa ngokushesha. Xhuma isikhombimsebenzisi sokuxoxa kwi-database engaxoshiwe, futhi ngeke nje usheshise ukufinyelela – uzosheshisa imiphumela.
3. Ubunjiniyela bomongo buba isiyalo esisemqoka
Izindleko zokufaka zamamodeli we-AI zilinganiselwa ku-300-400x ezinkulu kunokuphuma. Uma idatha yakho yomongo ihlelwe ngezinkinga ezifana ne-metadata engaphelele, i-HTML engasusiwe, noma i-vector engenalutho, iqembu lakho lizobhekana nokuwohloka kwezindleko ezinkulu ngenkathi licubungula esikalini. Ngaphezu kwalokho, okudidekile noma umongo ongaphelele futhi kuyinkinga enkulu yokwethenjwa kwe-AI, namagama omkhiqizo anqabile kanye nokubuyisa okuthe xaxa okudidayo ngenkathi izinguquko ezincane ziye ekuphumeni noma kumamodeli ahlukahlukene angaholela ekuphumeni okuhlukile okuhlukile.
Okungenza kukumangaze ukuthi ubunjiniyela bemongo buye ngaba yigama le-buzz buzz yedatha + ye-AI maphakathi nonyaka 2025. Ubunjiniyela bomongo buyinqubo ehlelekile yokulungiselela, ukwenza kahle, kanye nokugcina idatha yomongo yamamodeli we-AI. Amaqembu akwazi ukubheka phambili kokuqapha umongo wokuqapha-uqinisekisa i-Corpus ethembekileyo kanye nokushumeka ngaphambi kokushaya imisebenzi ebizayo yokucubungula – kuzobona imiphumela engcono kakhulu evela kumamodeli abo we-AI. Kepha ngeke kusebenze esilisweni.
Iqiniso ngukuthi ukubonakala kudatha yedatha elokuphela kukodwa akukwazi ukubhekana nekhwalithi ye-AI – futhi kungenjalo izixazululo zokubona ze-AI ezifana nokuhlolwa. Amaqembu adinga indlela ephelele enikeza ukubonakala ku -nke indlela yokwenza phakathi kwa- ukukhiqizwa – kusuka kudatha yomongo kumodeli kanye nemiphumela yayo. Indlela yezobuchwepheshe yezobuchwepheshe ehlanganisa Idatha + AI ndawonye ukuphela kwendlela yokuthola i-AI ethembekile esikalini.
4
Umbiko wakamuva we-MIT uthe konke. I-AI inenkinga yenani. Futhi ukusolwa kuphumula – okungenani ngokwengxenye – neqembu eliphezulu.
“Sisenabantu abaningi abakholelwa ukuthi i-AI ingumlingo futhi uzokwenza noma yini ofuna ukuthi yenze ngayo.”
Leyo yisilinganiso sangempela, futhi ifaka indaba evamile yamaqembu e-AI + AI
- Isikhulu esingaqondi ubuchwepheshe obubeka phambili
- Iphrojekthi yehluleka ukuhlinzeka ngenani
- Umshayeli wendiza uyamangala
- Hlambulula bese uphinda
Izinkampani zichitha izigidigidi kubashayeli bezindiza ze-AI ngaphandle kokuqonda okucacile kokuthi i-AI izoqhuba umthelela kuphi noma kanjani futhi inomthelela obonakalayo ngokungasebenzi kokusebenza kwezindiza, kepha intshiseko ye-ai ilonke.
Ukuthola ukubaluleka kudingeka kube owokuqala, okwesibili, kanye nokwesithathu. Lokho kusho ukuhlomisa amaqembu wedatha + AI aqonda ubuchwepheshe kanye nedatha ezoyifaka amandla ngokuzimela ukubhekana nezinkinga zebhizinisi zangempela – kanye nezinsizakusebenza zokwenza lawo macala athembekile.
I-5. Ukuphula ikhodi kuma-ejenti vs.
Ngenkathi izifiso ze-agentic bebelokhu zivusa umshini we-hype ezinyangeni eziyi-18 ezedlule, impikiswano ye-semantic phakathi “ama-agentic ai” an “agents” ekugcineni aqhutshwa emaphethelweni ahlanganisiwe weSigaba Sokuphawula nge-LinkedIn kuleli hlobo.
Enhliziyweni yalo bhizinisi kungumehluko wezinto ezibonakalayo phakathi kokusebenza nezindleko zalawa maqhinga amabili abonakala ebonakalayo kodwa amangazayo.
- Ama-ejenti anenhloso eyodwa kukhona ama-WorkHorses ngemisebenzi ethile, echazwe kahle lapho ubukhulu bucacile futhi imiphumela iyabonakala. Basebenzise umsebenzi ogxile, ophindaphindayo.
- Ukuhamba komsebenzi we-Agentic I-Tackle messy, izinqubo zezinyathelo eziningi ngokuzephula zibe yizakhi ezilawulekayo. Icebo liphula izinkinga ezinkulu kwimisebenzi emincane amamodeli amancane angaphatha, bese usebenzisa amamodeli amakhulu ukuze aqinisekise futhi ahlanganise imiphumela.
Isibonelo, i-ejenti yokuxazulula inkinga kaMonte Carlo isebenzisa ukuhamba komsebenzi we-agentic ukuhlela ama-orchestrate amakhulu amakhulu ama-sub-agents ukuphenya izimbangela zedatha + zekhwalithi yekhwalithi ye-AI.
I-6. Ukushumeka ikhwalithi kusendaweni yokubuka nokubheka kulungile ngemuva kwayo
Ngokungafani nemikhiqizo yedatha yasendulo, i-AI ngezindlela zayo ezahlukahlukene ayinqumi ngokwemvelo. Okungenayo akunjalo okuphumayo. Ngakho-ke, ukudiliza ukuthi yikuphi ukubukeka okuhle okunje kulomongo kusho ukulinga hhayi nje imiphumela, kodwa futhi nezinhlelo, ikhodi, nokufaka okubophayo.
Ukushumeka kunguhlelo olunye olunjalo.
Lapho ukushumekela kuhluleka ukumela incazelo ye-semantic yedatha yomthombo, i-AI izothola umongo ongafanele noma ngabe i-vector database noma imodeli. Okuyinto ebaluleke kakhulu ukuthi kungani ukuncengelwa kwekhwalithi sekuyinto ebaluleke kakhulu embilweni-2025.
Amakhefu ashumeke njalo anezingqinamba zedatha eziyisisekelo: ama-array angenalutho, ubukhulu obungalungile, amanani e-vector owonakele, njll. Inkinga ukuthi iningi lamaqembu lizothola lezi zinkinga lapho impendulo ikhona ngokusobala ayilungile.
Ikhasimende elilodwa leMonte Carlo lithwebule inkinga kahle: “Asinakho ukuqonda kokuthi kukhiqizwa kanjani ukushumeka, ukuthi iyiphi imininingwane emisha, nokuthi kuhlobene kanjani nenqubo yethu esetshenziswayo. Ngabe kufanele siqale kabusha amamodeli ethu?”
Njengoba ubukhulu obukhulu bekhwalithi nokusebenza buqala ukugxila, amaqembu aqala ukuchaza amasu amasha okuqapha angasekela ukufakwa kwamaqembu ekukhiqizeni; kufaka phakathi izici ezinjengobukhulu, ukuvumelana, kanye nokuphelela kwe-vector, phakathi kwabanye.
7. Imininingwane ye-Vector idinga isheke elingokoqobo
Ama-vector databases awasha ka-2025. Yini okusha ukuthi amaqembu wedatha + AI aseqalile ukuqaphela leyo datha ye-vector abathembele kuyo kungenzeka ukuthi bathembeke njengoba becabanga.
Ezinyangeni zokugcina ezingama-24, imininingwane ye-vector (yiphi idatha yesitolo njenge-veectors enezici eziphezulu ethatha ingqalasizinda ye-de facto yezicelo ze-rag. Futhi ezinyangeni ezedlule, sebephenduke umthombo wamaqembu wedatha + AI.
Ukushumeka ukudonsa. Amasu okushintsha amasu aguqukayo. Amamodeli wokushumeka avuselelwa. Konke lokhu kuguqulwa kudala ukonakaliswa kokusebenza okuthule okuvame ukungaziwa njenge-halliagnased njenge-hallucinations – nokuthumela amaqembu phansi izimbobo zonogwaja ezibizayo ukuze uzixazulule.
Inselelo ukuthi, ngokungafani nolwazi lwendabuko ngokuqapha okwakhelwe ngaphakathi, amaqembu amaningi awelanga ukuvela okudingekayo ekusesheni kwe-vector, ukushumeka, kanye nokuziphatha kwe-ejenti ukubamba izinkinga ze-vector ngaphambi komthelela. Lokhu kungenzeka kuholele ekukhuphukeni kokuqalwa kwedatha yedatha ye-Vector, kanye nezinye izixazululo zokuphawula zokwenza ngcono ukunemba kwempendulo.
8
I-AI Modeling Ukusingathwa komhlaba ukuhlanganisa umhlaba wonke acwebezelayo: ama-databricks kanye ne-AWS Bedrock. Zombili amapulatifomu ayaphumelela ngokufaka amakhono we-AI ngqo kwingqalasizinda yedatha ekhona esikhundleni sokudinga amaqembu afunde izinhlelo ezintsha ngokuphelele.
Ama-Databricks awina ngokuhlanganiswa okuqinile phakathi kokuqeqeshwa ngemodeli, ukuthunyelwa, kanye nokucutshungulwa kwedatha. Amaqembu angakwazi amamodeli amahle epulatifomu efanayo lapho idatha yawo ephila khona, eqeda ubunzima bedatha ehamba phakathi kwezinhlelo. Ngaleso sikhathi, i-AWS Bedrock iphumelele ngokuphepha kobubanzi nokuvikeleka kwebanga lebhizinisi, inikezela ukufinyelela kumamodeli wesisekelo amaningi avela ku-anthropic, Meta, nabanye ngenkathi egcina izindinganiso eziqinile zokuphathwa kwedatha kanye nezindinganiso zokuhambisana.
Yini ebangela abanye ukuba bawele ngemuva? Ukuhlukaniswa nobunzima. Amapulatifomu adinga umsebenzi wokuhlanganiswa ngokwezifiso obanzi noma amaqembu aphoqa ukuthi amukele amathuluzi amasha aphelele alahlekelwa yizisombululo ezilingana nokuhamba komsebenzi okhona.
Amaqembu akhetha amapulatifomu e-AI asuselwa ekusetshenzisweni okulula kanye namakhono wokuhlanganiswa kwedatha kunokuba asebenze ngendlela eluhlaza. Abaphumelele bayaqonda ukuthi imodeli enhle kakhulu ayinamsebenzi uma inkimbinkimbi kakhulu ukuthi iphakame futhi ilondoloze ngokuthembekile.
9
I-Model Contector Protocol (MCP) ivele njenge- “USB-C-C-C yemodeli ye-AI” – I-Universal Standard evumela izinhlelo zokusebenza ze-AI ukuthi zixhumeke kunoma yimuphi umthombo wedatha ngaphandle kokuhlanganiswa kwedatha ngaphandle kokuhlanganiswa kwedatha ngaphandle kokuhlanganiswa kwedatha ngaphandle kokuhlanganiswa kwedatha ngaphandle kokuhlanganiswa kwedatha ngaphandle kokuhlanganiswa okwenziwe ngokwezifiso.
Esikhundleni sokwakha izixhumi ezihlukile zazo zonke imininingwane, i-CRM, noma i-API, amaqembu angasebenzisa i-protocol eyodwa ukunikeza ukufinyelela kwe-LLMS kukho konke ngasikhathi sinye. Futhi lapho amamodeli angadonsa emithonjeni eminingi yedatha ngaphandle, aletha izimpendulo ezinembile.
AbakwaMfuthomukeli Basekuqaleni bavele babika ukuncishiswa okukhulu ekuhlanganiseni okuyinkimbinkimbinkimbinkimbi yokuhlanganisa nobunzima ngokugxila ekusetshenzisweni kwe-MCP okusebenza kuyo yonke imininingwane yabo ye-Ecosystem yabo.
Njengebhonasi, i-MCP ibuye ibuye ibe bukhoma kanye nokugawula izidingo – Izidingo ezibalulekile ukuthunyelwa kwebhizinisi.
Kepha ungalindeli ukuthi i-MCP ihlale i-static. Abaholi abaningi bedatha kanye nabakwa-AI balindele umthetho wokuqukethwe kwe-ejenti (ACP) ukuba aqhamuke kungakapheli unyaka olandelayo, basingatha izimo ezinzima zokwabelana ngomongo. Amaqembu amukele i-MCP manje azobe esekulungele le nqubekela phambili njengoba kuvela okujwayelekile.
10. Idatha engahleliwe yigolide elisha (kepha ingabe igolide liyigolide?)
Izicelo eziningi ze-AI zincike kwidatha engahleliwe – njengama-imeyili, amadokhumenti, izithombe, amafayela alalelwayo, namathikithi okusekelwa – ukuhlinzeka ngomongo ocebile owenza izimpendulo ze-AI.
Kepha ngenkathi amaqembu ekwazi ukuqapha idatha ehlelekile ngamathuluzi asunguliwe, idatha engahleliwe isebenze isikhathi eside endaweni eyimpumputhe. Ukuqashwa kwekhwalithi yesikhathi yesikhathi kwendabuko akukwazi ukuphatha amafayela wombhalo, izithombe, noma amadokhumenti ngendlela efanayo amatafula amatafula edatha.
Izixazululo ezinjengokuqashwa kwedatha kukaMonte Carlo ezingahlelwanga zibhekana naleligebe labasebenzisi ngokuletha amasheke ekhwalithi ezenzakalelayo emibhalweni kanye nezinkundla zezithombe eqhweni leqhwa, ama-databricks, kanye ne-biobquery.
Ukubheka phambili, ukuqapha kwedatha okungahleliwe kuzoba njengezinga lekhwalithi yendabuko yendabuko. Izinhlangano zizosebenzisa izinhlaka eziphelele zekhwalithi eselapha yonke imininingwane – ehlelekile futhi engahleliwe – njengezimpahla ezibucayi ezidinga ukuqapha okusebenzayo kanye nokubusa.
Ngibheke phambili ku-2026
Uma u-2025 usifundise noma yini kuze kube manje, yithi amaqembu awine ne-AI akuzona ezabelwa amabhilidi amakhulu noma amademoni akhanyayo. Amaqembu awina umjaho we-AI amaqembu athola ukuthi angaletha kanjani okuthembekile, anakele futhi athembekile Ayi ekhiqizweni.
Abaphumelele awenziwanga endaweni yokuhlola. Zenziwe ezandleni zabasebenzisi bangempela. Hambisa izixazululo ze-AI ezingatholwa, futhi uzoletha inani le-AI elibonakalayo. Kulula kanjalo.
