Ekushiseni kokuhlanya kokuqala kwe-ChatGPT, ngathola umbhalo ovela kowayesebenza naye. Wayefuna ukwenza umbono ngami. Ehlale ejabulela ukubhebhana, saxhuma ucingo wabe eseqala ngokuthi “Uyakhumbula ukuthi wawuhlale ungicela ukuthi ngikudonselele ama-data? Kuthiwani uma ubungakwenza wena ngokwakho?” Ube eseqhubeka nokunginika umbono wokuthi izinkulungwane (amashumi ezinkulungwane?) zabanye abantu zazicabanga ngesikhathi esifanayo: Ama-LLM angasetshenziselwa umbhalo-kuya-SQL ukusiza abantu abancane bezobuchwepheshe baphendule imibuzo yabo yedatha.  Ngangingene shí kulo mbono, kodwa ngaphambi kokungena ekhanda kuqala, ngatshela uLei (manje oyi-CTO yami) ukuthi kufanele senze ukuqinisekiswa okuthile. Sathintana nabangane kanye nalabo esasisebenza nabo abavela ezimbonini ezihlukahlukene. Kube nentshisekelo eqinile "kukuhlaziya kokuzisiza." Besazi ukuthi kuzoba nzima kakhulu kunalokho obekubukeka, kepha ithuba lizwakala limnandi kakhulu ukuthi singaliyeka. Ngakho-ke mina noLei sasuka e-Shire futhi saqala uhambo lwethu lokudala umbono wethu:  .   Fabi.ai  Lokhu okuthunyelwe akuphathelene nomkhiqizo wethu ngokwawo (noma kunjalo, uma ufuna ukwazi, ungafunda kabanzi mayelana nokuthi eminye imibono engezansi yazisa kanjani umsebenzi wethu wakamuva womkhiqizo.  ). Kunalokho, bengifuna ukwabelana ngokufundiwe okubalulekile esikuqoqile ekusebenzeni nama-LLM ukuze sihlaziye idatha ohambweni lwethu.   lapha   Qaphela: Lolu hambo luntuleka ngokudabukisayo kubathakathi kanye nezimpi ezidumile ze-Middle-earth. 🧙  Kungani usebenzise i-AI ukuze uthole ukuhlaziya kokuzisiza?  Ngeke sihlale kokuthi “kungani” isikhathi eside kakhulu. Uma ufunda lokhu, kungenzeka uwele kwelinye lamaqembu amabili:  Ungumuntu ofisa ukuthi ube nezibalo zokuzisiza ezitholakalayo futhi ongafuni ukuthi uhlale ulindile ethimbeni lakho ledatha  Useqenjini ledatha futhi ubulokhu uzwa ukuthi i-AI izokusiza kanjani ukuxazulula inkinga yakho yezicelo zesikhashana.  Ukuziba ukukhathazeka kwendima yabahlaziyi bedatha nososayensi, umqondo we-AI ekwazi konke engaphendula noma yimiphi imibuzo mayelana nedatha yenhlangano uzwakala umuhle. Noma okungenani, kuzwakala kumnandi enhlanganweni nakubaholi bayo bebhizinisi abanobuhlakani bezindlela ezintsha zokubuza imibuzo abunamkhawulo. Le AI ingaba yisixazululo sokudala inhlangano “eqhutshwa yidatha” lapho wonke umholi encike ebufakazini obunamandla ukuze enze izinqumo zabo zamasu. Futhi konke ngengxenyana yezindleko obekuzovame ukuzithatha. Ekugcineni! Izinhlangano zingasebenzisa lawo “mafutha amasha” ezilokhu zizwa ngawo kusukela ngo-2010.  Kepha uma lokhu kuyinkinga ebaluleke kangaka okufanele ixazululwe futhi i-AI isibe yinhle kakhulu, kungani ungekho umkhiqizo   kuze kube manje? oyixazulule  Kungani i-AI ye-self-service analytics yehlulekile kuze kube manje  Inhlolovo yemboni yakamuva ipenda isithombe esiyinkimbinkimbi sokutholwa kwe-AI ebhizinisini.  bazama ama-agent e-AI. Nokho, abaningi bakhathazeka ngokwethembeka nokuvikeleka. Eqinisweni, u-21% wezinhlangano aziwasebenzisi nhlobo. Lokhu kunqikaza kuzwakala kakhulu emaqenjini edatha, lapho ukunemba nokwethembeka kuwumgomo obalulekile ekhonweni lethu lokwenza umsebenzi.   Amaphesenti angama-61 ezinkampani  Abamukeli be-AI-ikakhulukazi ebhizinisini-banebha ephezulu uma kuziwa kulokho okulindelekile kobuchwepheshe. Kumongo wokuhlaziya idatha kanye nephupho lokuzisiza, silindele ukuthi ithuluzi lethu le-AI:  Inikeza imininingwane: Amathebula namashadi mahle, kodwa lawo ayisethi engaphansi yalokho umuntu angakubiza ngokuthi “imininingwane”. Imininingwane ithi “Aha!” izikhathi ezivela ekuboneni izinto kudatha yakho eziphikisana nokuzwa kwakho futhi bezingeke zicatshangelwe ngenye indlela. Kwesinye isikhathi umbuzo we-SQL noma i-pivot ingakhanyisa le mininingwane, kodwa ngokuvamile izwakala kufana nokuthola inaliti ku-haystack.  Sebenza ngokuthembekile cishe u-100% wesikhathi: Okuwukuphela kwento embi kakhulu kunokungabi nadatha idatha embi. Uma i-AI ingenakuthenjwa noma iveza izimpendulo nedatha, lokho kupela izindaba ezimbi zawo wonke umuntu. Lokhu kusho ukuthi uma i-AI inedatha, kufanele iyisebenzise ngendlela efanele. Kodwa uma ingenayo idatha, kufanele igweme ukunikeza impendulo (into ama-LLM adume kabi ngayo).  Finyelela ezinhlobonhlobo zamasethi amakhono obuchwepheshe: Ubuhle bama-LLM ukuthi ungakwazi ukusebenzelana nawo njengoba ubungenza nozakwenu nge-Slack. Ungasebenzisa ulimi olungacacile. Omunye umuntu noma into kungenzeka iqonde isicelo sakho kumongo webhizinisi. Ngokuphambene, lapho isistimu idinga kakhulu ukusebenzisa amagama aqondile ngendlela eqondile, yilapho ifinyeleleka kancane. Lolu hlobo lwesistimu ludinga ukuqeqeshwa nokuqiniswa, okuyinto sonke esiyaziyo, kungaba inselele.  Ngokudabukisayo, izixazululo eziningi zamanje zisebenzisa uhlaka lwendabuko lwe-monolithic AI, oluvame ukwehluleka ukuhlangabezana nokulindelwe. Eminyakeni embalwa edlule, ithimba le-Fabi.ai kanye nami sasebenza kanzima kulolu daba. Sakhe ama-prototypes ebhizinisi futhi sahlola izinketho eziningi. Ekugcineni, saqaphela ukuthi akukho Retrieval Augment Generation (RAG) noma ukulungisa kahle okungalungisa le nkinga ngohlaka lwamanje lwe-monolithic.   Lapho sihlola le ndlela, izinto ezimbalwa zasicacela:  I-RAG ithambile. Umongo omncane kakhulu futhi i-AI ayikwazi ukuphendula umbuzo futhi ibeka engcupheni yokubona izinto ezingekho. Umongo omningi futhi i-AI iyadideka futhi ilahlekelwe ukunemba kwayo.  I-AI yeshothi eyodwa ayikufikisi ndawo. I-AI ehamba phambili emhlabeni ayisoze yakwazi ukudonsa nokuhlaziya idatha ngokudutshulwa okukodwa. Kunama-nuances amaningi kakhulu kudatha nasembuzweni. Ake sithathe isibonelo esilula ngangokunokwenzeka: Unenkambu ethi “Uhlobo lwe-akhawunti” enamanani angu-95% ahlukene angu-10. Uma ucela i-AI ukuthi ihlunge kusethi yezinhlobo ze-akhawunti, ingase yehluleke ukubona ukuthi kunamanani angenalutho, ngaleyo ndlela ikhiqize umbuzo we-SQL ongavumelekile. “Impela,” ungase uthi, “kodwa singavele sibale izibalo zenkambu ngayinye namavelu esampula bese sikugcina esitolo sethu se-vector.” Izinhlobo zezinkinga cishe azinamkhawulo futhi zonke zihlukile ngendlela yazo.  Idatha yebhizinisi ingcolile. Lokhu kuhlobene namaphuzu amabili okuqala, kodwa kufanele kugcizelelwe. Noma ngabe okomzuzwana nje, inhlangano ingaba namatafula ambalwa egolide anesendlalelo se-semantic esichazwe kahle, wonke avele aphahlazeke ngokushesha nje lapho umholi we-RevOps enquma ukulungisa imodeli yebhizinisi. Ngithanda ukudweba isifaniso sendlu: Ungakwazi ukugcina indlu icocekile, kodwa kuhlale   okudinga ukuhlanzwa noma ukulungiswa. kukhona  Umbhalo-kuya-SQL unomkhawulo kakhulu. Emibuzweni eminingi yokuhlaziya idatha, ukubhala i-SQL ukuze udonse idatha kuyisinyathelo sokuqala nje. Lesi yisinyathelo okufanele usithathe ngaphambi kokuthi uqale ukubuza imibuzo ethokozisayo. I-SQL ayikwazi nje ukusingatha ukuhlaziya okuyinkimbinkimbi abasebenzisi bebhizinisi abakubuzayo. Ngakolunye uhlangothi ama-LLM kanye nePython awufanele ngokuphelele umsebenzi. Lawa mathuluzi angathatha okukhiphayo kwakho kwe-SQL futhi athole leyo naliti ku-haystack. Bangaphinda benze ukuhlaziya kokuhlehla ukuze bathole amathrendi amakhulu.  Ngemuva kokubheka lezi zinkinga, sicabange ukuthi singayenza kanjani i-AI ivumelane nezinkinga. Kulapho ama-agent e-AI aqala khona ukudlala futhi asiqinise lo mqondo.  Ikusasa: Ama-agent meshes  Emzuzwini lapho sibheka izinhlaka ze-agency, sasazi ukuthi zizoshintsha umdlalo. Ngokungazelelwe saba nomuzwa wokuthi singavumela i-AI inqume ukuthi ingayiphendula kanjani imibuzo. Ingasebenza ngezinyathelo futhi ixazulule ngokwayo. Uma i-AI ibhala umbuzo we-SQL ogeja amanani angenalutho kunkambu ethi "Uhlobo lwe-akhawunti", ingakwazi ukuqalisa umbuzo, ibone iphutha, futhi izilungise yona ngokwayo. Kepha kuthiwani uma singathatha lesi sinyathelo siqhubeke futhi sivumele i-AI ukuthi isebenze kakhulu ePython futhi isebenzise ama-LLM? Manje, i-AI yenza okungaphezu kokudonsa idatha. Ingasebenzisa amaphakheji e-Python noma ama-LLM ukuze uthole izinto eziphuma ngaphandle, amathrendi, noma imininingwane ehlukile, ngokuvamile okuzodingeka uyibheke ngesandla.  Kodwa sasisenenkinga eyodwa: idatha yebhizinisi engcolile. Sikholelwa ukuthi izinhlangano zingaxazulula lokhu ngokusebenzisa izinqubo eziqinile zobunjiniyela bedatha, njenge-a  kanye nongqimba oluqinile lwe-semantic. Kodwa-ke, asivamile ukuthola izinhlangano ezenza lokhu empilweni yangempela. Izinhlangano eziningi zisebenzisa amaspredishithi, amathebula abhakwe uhhafu, kanye namamodeli edatha ashintsha njalo. Kusukela lapha, siqhamuke nombono wokwakha ama-agent akhethekile e-AI angakhiwa ngokushesha ukuze aphendule isethi ethile yemibuzo.   medallion bokwakha  Njengoba izinkampani zikhula, ziphatha idatha eyengeziwe futhi zinabasebenzisi abaningi. Umbono wemeshi yomenzeli usiza ukulinganisa ukwenza izinqumo okusheshayo nokulawula okudingekayo ekubuseni. Ama-ejenti akhethekile asiza ukubeka imingcele ecacile nezibopho ze-AI ngayinye. Baphinde badale indlela enwebekayo yokuthi ama-ejenti axhumane. Futhi, bangasiza ukuphatha izinsiza ngendlela efanele kuwo wonke amaqembu nezinkampani.  Ama-agent akhethekile e-AI  Umqondo we-ejenti ekhethekile ukuthi lo menzeli angakwazi futhi uzophendula kuphela imibuzo kudathasethi echazwe ngokuqinile. Isibonelo, ungakha futhi uqalise i-ejenti ye-AI ephendula imibuzo mayelana nemikhankaso yokumaketha. Noma ungakha enye ukuze uphendule imibuzo mayelana nepayipi lokumaketha, njalo njalo njalo.   , usebenzisa lesi sakhiwo. Izimpawu zakuqala zithembisa kakhulu. Uma amasethi edatha ekhethwe ngokucophelela futhi esezingeni elifanele lembudumbudu, lawa ma-ejenti angaphendula isethi ethile yemibuzo ngokwethembeka ngokwedlulele. Umakhi walawa ma-ejenti angabelana ngawo nabasebenzisi abangebona abezobuchwepheshe futhi aphumule kalula azi ukuthi i-AI ngeke iphendule imibuzo engekho endaweni.   Sisanda kwethula i-Agent Analyst  Kukhona iphutha elilodwa kuphela: Abasebenzisi badinga ukwazi ukuthi iyiphi i-ejenti okufanele baye kuyo kumuphi umbuzo. Kufana nokudinga ukwazi umhlaziyi wezokuthengisa olungile ukuze abuze umbuzo wokuqhathanisa nokubuza nje umbuzo ojwayelekile. Ngombuzo ojwayelekile, othile eqenjini angawuqondisa kumuntu ofanele. Yilapho umqondo "we-agent mesh" ungena khona.  Ama-ejenti wokuxhuma ndawonye  Uma umenzeli oyedwa ekwazi ukuphendula imibuzo eqondene nesizinda ngokuthembekile, kungani-ke ungavumeli ama-ejenti akhulume ngomunye nomunye? Kungani, isibonelo, i-ejenti yomkhankaso wokumaketha ingakwazi ukubuza i-ejenti yamapayipi ngokuqondile uma ikwazi ukuphendula umbuzo kalula? Sikholelwa ukuthi kufanele ikwazi. Eqinisweni, sicabanga ukuthi esikhathini esizayo kuzoba namanethiwekhi ama-agent anesakhiwo se-hierarchical. Ungathatha isithombe “kumenzeli we-GTM” obiza “i-ejenti yokumaketha.” Lo menzeli ube esebiza kokubili “umenzeli wepayipi” kanye “Ne-ejenti yomkhankaso wokumaketha.”  Lo mbono ufana nombono ojwayelekile ozungeza i-AI owaziwa ngokuthi "  ." Kuyikusasa lapho abenzeli be-AI besebenzisana ngokushelela kuzo zonke izinhlangano. Benza lokhu ngenkathi beqinisekisa ukuthi ukuvikeleka nokwethenjwa kuhlala kuqinile.    I-inthanethi yama-ejenti  Le ndlela ye-mesh inikeza izinzuzo ezimbalwa ezibalulekile ngaphezu kwe-monolithic AI (kusendlalelo se-pristine semantic):  Ukuqaphela: Njengoba umenzeli oyedwa enikeza izimpendulo ezisuselwe kudatha ethile, ungakwazi ukulandelela impendulo ngayinye ubuyele kulowo menzeli. Lokhu kusiza ukuqinisekisa ukunemba ngokucwaninga. Ukuze unikeze ukhonkolo, noma olulula kakhulu, isibonelo, zicabange unamathebula emicimbi amabili: elokuthengisa elinye elomkhiqizo. Uma umsebenzisi ebuza, “Imiphi imicimbi engenise imali engenayo eningi?” i-AI ingase icabange ukuthi isho imicimbi yomkhiqizo. Ngisho noma kungalungile, umsebenzisi angabona ukuthi yimuphi umenzeli ophendulile futhi angaqondisa i-AI.  Ukulondolozeka: Njengenjini yemoto, uma ungathola kalula izinkinga futhi ushintshe izingxenye ngokushesha, imoto iba qotho kakhulu. Uma i-ejenti eyodwa iqala ukwehluleka ngenxa yoshintsho kumodeli yedatha, ingabonwa ngokushesha futhi lowo menzeli angabuyekezwa.  Ukunemba: Nge-ejenti ngayinye esebenza ngaphakathi kwemingcele yayo, asikho isikhala sokuthi iphume kojantshi. Ingase ingabi nayo impendulo, kodwa ngeke yenze okuthile okujabulisayo.  Ekupheleni kosuku, lo mbono we-mesh awuyona inoveli. Lokhu kufaka izibuko umqondo wengxube yochwepheshe ekhonjiswe ukuthuthukisa ukunemba kwama-LLM. Kumane kuthathe lowo mbono ofanayo futhi ulethe kubasebenzeli be-AI.  Izinselele zobuchwepheshe zama-ejenti anezikhala  Kwa-Fabi.ai, luselude ukhalo okufanele siluhambe njengoba sakha i-Analyst Agent mesh. Kodwa, sesivele sizinqobile ezinye zezinselele ezinkulu zengqalasizinda yezobuchwepheshe.  Ama-ejenti womhlaziyi wedatha we-AI adinga ukwakheka okuhlukile. Lo mklamo kufanele ubavumele ukuthi basebenzise i-Python noma i-LLM ukuze baphendule imibuzo, bahlale bevumelanisa nemithombo yedatha, futhi bangene ezinkundleni ezihlanganyelwe, kuyilapho zisahlezi zivikelekile futhi zinyuka. I-ejenti ngayinye idinga ukusebenza nge-Python kernel yayo, edinga ukuphothwa ngokushesha phezulu noma phansi ukuze kuncishiswe izindleko futhi ihlale ivunyelaniswa nedatha yomthombo.   Izakhiwo ezingahlinzeki ngezinhlamvu ezingazodwana kumenzeli ngamunye zingangena kweyodwa yalezi zingozi ezilandelayo:  Ukungqubuzana kombuso kokuguquguqukayo phakathi kwama-agent we-AI. Ama-ejenti amabili ahlukene angase akhiqize okuguquguqukayo kokuthi “foo” ukuze aphendule umbuzo, abangele ukungqubuzana. Kungase kube nezinye izindlela zokukhiqiza izihlonzi ezihlukile, kodwa lezi zandisa amathuba okuthi i-AI ikhiqize ikhodi engavumelekile.  Izingozi zokuphepha ezibangelwa ukwabelana kwedatha phakathi kwamaqembu ahlukene noma ngisho nezinhlangano ezihlukene.  Ukusebenza kuba nomthelela uma i-ejenti eyodwa ithatha inani elingalingani lezinsiza zokubala.  Inselele yokwakha lolu hlobo lwenkundla iyinselelo ye-AI njengoba kuyinselelo ye-DevOps.  Ukubheka phambili: Ukwamukela ama-agent akhethekile, alawulwayo e-AI kudatha  Njengoba izinkampani zamabhizinisi zilawula izinhlelo zokusebenza eziningi ze-AI ekusebenzeni kwazo, zidinga izindlela ezikhethekile nezilawulwa kahle. Uhlaka lwemeshi yomenzeli lusebenzisa ama-ejenti wedatha we-AI akhethekile njengendlela yokukala i-AI ekuhlaziyeni idatha. Le ndlela igcina ukuphepha, ukwethembeka, nokusebenza kuqinile.  Besingahle silindele ukuthi i-AI ibe yonke indawo manje, iphendula imibuzo eminingi yedatha. Kodwa, uma sibhekisisa kahle, inqubekelaphambili eminyakeni emibili nje selokhu kwethulwa i-ChatGPT iyamangalisa. Kusekuningi okumele sikufunde kulolu hambo. Emqondweni wami, nokho, ama-ejenti kanye nezinhlaka ze-agent mesh kuzoba ukhiye ebhizinisini le-AI.

This story contains new, firsthand information uncovered by the writer.

The is an opinion piece based on the author’s POV and does not necessarily reflect the views of HackerNoon.

This post provides insights into new product. 

Lo msindo ukhiqizwa ngolimi lokuqala lwendaba!

I-Text-to-SQL Bekufanele Kube Uhlelo Lokubulala Lwe-AI. Akunjalo.

About Author

IMIBONO

HANG TAGS

LESI SIHLOKO SETHULWE NGAPHAKATHI

Related Stories

Zulu Network: Moving The Bitcoin Economy Forward With a Two-Tiered Bitcoin Layer 2 Architecture

The Billionaires F*cked Up

Zulu Network: Moving The Bitcoin Economy Forward With a Two-Tiered Bitcoin Layer 2 Architecture

The Billionaires F*cked Up

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps