Amaphutha Athule Wokushumeka: Kungani I-AI Yakho Ikuthola Okungalungile
Idatha yombhalo ongahlelekile yandile kamuva nje, kusukela ekubhalweni kwezobuchwepheshe nezincwadi zesayensi kuya kokuthunyelwe kwezokuxhumana nokuhlaziya amakhasimende. Amabhizinisi avela emikhakheni eyahlukene aya ngokuya aqaphela ukuthi lo mbhalo unikeza imininingwane eyigugu, kodwa ukuthola, ukuhlela, nokusebenzisa le datha kunzima.
Ukunotha kwe-semantic yolimi lwabantu akuthathwanga ngamasu avamile wokucubungula umbhalo asekelwe egameni elingukhiye, abonise ukuthi ayanele. Isibonelo, ukusesha "izinkinga zemoto" kungase kungavezi amaphepha abalulekile amayelana "nezinkinga zezimoto", futhi isibuyekezo somkhiqizo esikhuluma "ukufriza kwesikrini" singase singaphuthelwa lapho kuhlolwa izikhalazo zamakhasimende mayelana "nokuphahlazeka kwesistimu."
Ukushumeka umbhalo, indlela yenoveli eguqula amagama nemisho ibe amavekhtha ezinombolo athwebula incazelo yawo, kwathuthukiswa futhi kwasetshenziswa ngenxa yalesi sibopho esiyisisekelo. Amafemu amakhulu ezobuchwepheshe achithe izigidigidi ekudaleni amamodeli okushumeka athuthuke kakhulu; Ukushumeka kwe-OpenAI, i-Meta's RoBERTa, i-BERT ye-Google, kanye nenqwaba yezinye izindlela zomthombo ovulekile manje kuyizingxenye ezibalulekile zezinhlelo ze-NLP zesimanje.
Nokho, naphezu kokusetshenziswa kwawo kakhulu, asikaqondi ngokugcwele ukuthi lawa mamodeli wokushumeka asebenza kanjani kuzilungiselelo ezingokoqobo. Lokhu kuntuleka kokuqonda kube nomphumela wokuthi:
- Amaphutha okusebenzisa abizayo avela lapho amasistimu angazanelisi izidingo zebhizinisi.
- Okuhlangenwe nakho okungekuhle komsebenzisi lapho okuqukethwe okuhlobene kugejwa ukusesha noma izinhlelo zokuncoma. Umehluko wokusebenza phakathi kwezibalo zabantu abahlukahlukene, izigaba zokuqukethwe, nezilimi.
- Ukusetshenziswa okungasebenzi kahle kwezinsiza yizinhlangano ezisebenzisa amamodeli ayinkimbinkimbi kakhulu
Ukufaneleka Nokusebenziseka Embonini
Izimboni eziningi ezahlukene zingazuza ngokuqondile kulokhu kuhlaziya:
Ukudayisa kanye ne-E-Commerce:
- Ukuthuthukisa usesho lomkhiqizo ukuze kuhlangatshezwane nemibuzo yezilimi eziningi, amagama afanayo, kanye nokungapeli kahle kahle.
- Ukuthuthukisa izinhlelo zokuncoma ukuhlonza imikhiqizo eqhathanisekayo enezincazelo ezihlukene.
- Ukuthuthukisa amathuluzi okuhlaziya athuthuke kakhulu awazi ubuhlakani bemizwa.
Ukunakekelwa Kwezempilo:
- Ukuvumela amagama ezokwelapha ukuthi amataniswe kukho konke ukuhluka kwamanothi kumasistimu okubuyisa amadokhumenti omtholampilo.
- Ngaphandle kokuthola imfundo, ukuthuthukisa ukucutshungulwa kolimi lwesiguli ukuze kuchazwe izimpawu.
- Ukuthuthukisa usesho lwezincwadi zezokwelapha ngokufanisa kangcono imiqondo kuzo zonke izinto eziguquguqukayo zobuchwepheshe
Ezezimali:
- Ukuthuthukisa ukuqapha ukuthotshelwa kokwephulwa kwenqubomgomo, ngaphandle kokubheka amagama.
- Ukuthuthukisa ukutholwa kokukhwabanisa ngokuhlonza amaphethini angabazekayo kuzo zonke izinhlobo zezilimi.
- Ukuthuthukisa isevisi yamakhasimende ngokuqonda kangcono izinhlobo ezahlukene zemibuzo yabathengi
Ukuthola ukuthi imodeli yakho yokushumeka ibuyisa kanjani imiphumela efanayo: Lokhu kuhlaziya kusiza izinhlangano zenze izinqumo ezingcono mayelana nokukhetha imodeli, izidingo zokucubungula kusengaphambili, namasu okuthuthukisa okungenzeka ngokunikeza indlela ehlelekile yokuqonda ukuziphatha kokushumeka. Le ndlela ekugcineni iholela ezinhlelweni zokucubungula ulimi ezisebenza kahle, ezithembekile, nezilinganayo.
Kuyini Ukushumekwa Kombhalo?
Kwabangaqaliwe, ukushumeka kombhalo kuguqula amagama noma imisho ibe amavekhtha ezinombolo - ikakhulukazi izinhlu ezinde zezinombolo, ama-vector aminyene anezilinganiso eziningi. Lawa ma-vector abekwe endaweni enobukhulu obuphakeme lapho ukufana kwe-semantic kuvezwa ukuba seduze. Ngamagama alula, izinto ezinezincazelo ezifanayo kufanele zisondelane ndawonye kulesi sikhala. Lokhu kushumeka amandla cishe kuyo yonke into ku-NLP yesimanje:
Uma usesha i-Google, iyaqonda ukuthi ubuqonde "amahhotela aseNew York" noma ngabe uthayiphe "izindawo zokuhlala e-NYC."
Uma iklayenti lakho le-imeyili liphakamisa izimpendulo
Uma isistimu yezincomo zokuqukethwe ithola izindatshana ongazithanda
Leli khono linika amandla izinhlelo zokusebenza eziningi:
- Izinjini zokusesha ze-Semantic
- Amasistimu wokuncoma okuqukethwe
- Ukuhlonza inhloso ye-chatbot
- Ukuhleleka nokuqoqwa kwemibhalo
- Izinhlelo zokuhlukanisa imibhalo kanye nokuphendula imibuzo
Ukusebenza kwalezi zinhlelo zokusebenza kuncike kakhulu ekutheni imodeli yokushumeka ikuphatha kahle kangakanani ukwahluka kwemibhalo ehlukahlukene kanye nokuvela kolimi. Nakuba imodeli ezwela kakhulu ezinhlobonhlobo ezincane zokufometha ingaba nenkinga ngemibuzo yosesho yangempela, imodeli ephatha "Ngiyawuthanda lo mkhiqizo" kanye nokuthi "Ngiyawuzonda lo mkhiqizo" njengoba icishe ifane ngeke isebenze ekuhlaziyeni imizwa.
Kungani Ngiqale Ukuphenya Amamodeli Okushumeka
Angisoze ngalukhohlwa usuku lapho ngangicwaninga amamodeli wokushumeka kwelinye lamakhasimende ami ngamasampula amadokhumenti. Ngizamile ukusesha "amakhompyutha aphathekayo angenazo izikrini zokuthinta," ukuze ngithole impendulo mayelana namakhompyutha aphathekayo wesikrini esithintwayo kunkomba ye-vector. Imodeli yethu yokushumeka yayiphuthelwe ngokuphelele ukuphika.
Lowo mzuzu wanamathela kimi. Ngisho, uma imodeli ingakwazi ukuqonda okuyisisekelo "ngaphandle", yini enye eyayishoda? Leso siqephu sangiholela endleleni engangingayilindele. Kwadingeka ngazi ukuthi inkinga yayibucayi kangakanani. Impela, amabhentshimakhi emodeli yethu yokushumeka abukeka emangalisa, kodwa aziphathe kanjani ngolimi olungcolile, lomhlaba wangempela abasebenzisi bethu abaluthayipha nsuku zonke? Kwenzekani lapho amagama angapelwanga kahle, imisho ingaphelele, noma abantu basebenzisa ama-emoji?
Ngemva kweminyaka emibili yokuzama ngamamodeli ahlukahlukene okushumeka kuzo zonke izinhlelo zokusebenza kusetshenziswa i-RAG Experiment Accelerator ( https://github.com/microsoft/rag-experiment-accelerator ), ngithuthukise indlela ehlelekile yokuhlola ukuthi lawa mamodeli aphatha kanjani izinhlobo ezahlukene zokuhlukahluka kwemisho. Engakuthola kwangethusa.
Lokhu kwakungewona amaphutha angahleliwe - kwakuyizindawo eziyimpumputhe ezijulile, ezihlelekile ezisekelwe endleleni lezi zinhlobo ezibona ngayo ulimi. Futhi akekho owayekhuluma ngakho. Ngabelana ngalokhu okutholiwe ngoba ngibone amaqembu amaningi echitha izinyanga esebenzisa amasistimu agxile ekushumekeni athole sekudlule isikhathi ukuthi agqekeza ngezindlela ezingalindelekile.
Lolu daba akulona nje lwezemfundo. Imiphumela yokungaqondi kahle ukuziphatha kokushumeka mibi: Osebenza naye enkampanini enkulu yezokunakekelwa kwempilo wangitshela ukuthi uhlelo lwabo lokubuyisa ulwazi lwezokwelapha lushoda ngamadokhumenti omtholampilo afanelekile ngenxa yokuthi imodeli yabo yokushumeka ayikwazanga ukuphatha kahle izifinyezo kanye nokwehluka kwamagama ezokwelapha. Ekunakekelweni kwezempilo, lolu hlobo lokugeja lungathinta izinqumo zokwelashwa.
Enkampanini yezinsizakalo zezezimali, isistimu yabo yokuqapha ukuthobela igeja ukwephulwa kwenqubomgomo ngenxa yokuthi imodeli yabo yokushumeka ayikwazanga ukubona ukuthi imisho yezwi yokwenziwa ("imali idluliselwe") isho into efanayo nezwi elisebenzayo ("umuntu odlulise izimali"). Ngike ngabona izinkampani ze-e-commerce zilahlekelwa izigidi zamaRandi ngenxa yokwehluleka kohlelo lwazo lokusesha imikhiqizo ukwamukela ama-typos ajwayelekile nezinqamuleli zolimi ezisetshenziswa abathengi.
Indlela Yami Yokuhlola
Ngithuthukise uhlaka lokuhlola oluhlola ukuthi amamodeli wokushumeka asabela kanjani ezigabeni eziningi ezihlukene zokuhlukahluka kombhalo. Ngisebenzisa ukufana kwe-cosine njengemethrikhi (lapho u-1.0 esho incazelo efanayo futhi u-0.0 esho okungahlobene ngokuphelele), ngisebenzise amakhulukhulu ezigameko zokuhlola. Ngizohamba kulokhu engikutholile, ngigxile kumamodeli amaningi, okuhlanganisa i-MSMarco DistilBERT, ukushumeka kombhalo we-OpenAI, nokunye okuningi, futhi ngibone amaphethini afanayo kukho konke ukushumeka okusekelwe ku-transformer.
Angikwazi Ukusho Okuphezulu Kusuka Kuzinhlamvu Ezincane (Nakancane)
Lokhu kungithusile. Amamodeli wokushumeka abona " i-Apple yamemezela imikhiqizo emisha" kanye "ne-apula lamemezela imikhiqizo emisha" njengento efanayo KANYE - amaphuzu afanayo ayi-1.0. Awukho umehluko.
Ngigijime kulokhu ngesistimu yekhathalogi yomkhiqizo. Ukusesha akukwazanga ukuhlukanisa phakathi komkhiqizo "i-Apple" nesithelo "i-apula." Amakhasimende asesha imikhiqizo ye-Apple athole izindlela zokupheka zikaphayi wama-apula. Ucabanga ukuthi amakhasimende akho azojabula ngokubuka imiphumela? Ngiqinisekile ukuthi ngeke ngikuthande.
Kungani lokhu kubalulekile? Cabanga ngazo zonke izimo lapho osonhlamvukazi kushintsha okusho – "Polish" vs "polish", "March" vs "mashi", "Bill" vs "bill". Emibhalweni yezomthetho noma yezokwelapha, lokhu kwehluka kungabaluleka kakhulu. Sisebenza namamodeli angaboni kancane kubo bonke isici solimi olubhaliwe. Kukhona indlela yokulungisa lezi zinhlobo zezinkinga, futhi zizodingidwa kamuva. Masiqonde izindaba ezengeziwe ngokushumeka. Qaphela ukuthi lokhu kungaba inzuzo uma lo mehluko ungathinti icala osiphethe.
Izinombolo Zingahle Zenziwe
Nalokhu kwangiphatha kabi. Amamodeli okushumeka abona "I-investimenti ibuyise u-2% ngonyaka" kanye nokuthi "Utshalomali lubuyise u-20% ngonyaka" njengento efanayo, amaphuzu aphezulu afanayo angu-0.97. Awukho umehluko obonakalayo phakathi kwalezi zimo ezimbili.
Ngihlangabezane nalokhu ngohlelo lokusesha amadokhumenti ezezimali. I-algorithm ayikwazanga ukubona umehluko phakathi "kwemali yokuphatha: 0.2%" kanye "nemali yokuphatha: 2.0%. Abatshalizimali abafuna izimali ezishibhile bathola izinketho ezibizayo ezinconyiwe. Ucabanga ukuthi i-akhawunti yakho yomhlalaphansi ingalijabulela lelo phutha? Ngiqinisekile ukuthi ngeke ngithande ukubona ukonga kwami kudliwa yizindleko.
Cabanga ngazo zonke izimo lapho amanani ezinombolo eyimiyalelo yomthamo obalulekile, ukubekezelelwa kobunjiniyela, imbuyiselo yezimali, kanye nezinsuku zokugcina zenkontileka. Ngombhalo wokutshalwa kwezimali noma wezokwelapha, lokhu kwehluka kungashintsha impilo. Empeleni sisebenza namamodeli angakwazi ukufunda ngezinombolo naphezu kokuphatha umbhalo ogcwele amanani abalulekile.
Inkinga ethi "Hhayi" Iyesabeka
Lokhu kuyingozi ngempela. Ukwengeza okuthi "hhayi" emshweni - ukuguqula ngokwezwi nezwi incazelo yakho - akuthinti kakhulu amaphuzu afanayo. Ngokuvamile sasibona amaphuzu angaphezu kuka-0.95 ngokuphikisana okuphelele. "Ukwelashwa kwathuthukisa imiphumela yesiguli" vs. "Ukwelashwa akuzange kuthuthukise imiphumela yesiguli" → 0.96 ukufana Lapho ngibonisa lokhu udokotela owayesebenzisa uhlelo lwethu lokusesha lwezokwelapha, wethuka. Wethuka waze wahlehla ekhompyutheni. Futhi kufanele abe.
Sasakha uhlelo odokotela ababezolusebenzisa ukuze bathole imigomo yokwelashwa. Yenza lokhu okungalungile, futhi abantu bangafa. Ukuphika akulona icala - kuyisisekelo olimini lwabantu. Uma usesho lwakho, isincomo, noma isistimu yokuhlaziya ingakwazi ukuhlukanisa "okusebenzayo" kokuthi "okungasebenzi" noma "kuphephile" kokuthi "okungaphephile", wakha imishini eyingozi yokubona izinto ezingekho.
Ekunakekelweni kwezempilo , kungasho ukuncoma ukwelashwa okuyingozi. Emadokhumeni omthetho , kungase kuguqule ngokuphelele izibopho zenkontileka. Ekulinganisweni kokuqukethwe , ungase uphuthelwe umehluko phakathi kokuthi "udlame luyamukeleka" nokuthi "udlame alusoze lwamukeleka."
Izikhala Azinandaba (Kuze Zenze Ngempela)
Izikhala ezengeziwe, amathebhu, ukufometha okuyinqaba - amamodeli awanandaba. Ukufana kuhlala ngaphezu kuka-0.995. Kodwa susa zonke izikhala? Ukufana kwehlela ku-0.82 kungazelelwe. Ngihlangabezane nale nkinga ngenkathi ngisebenza nokuqukethwe okuklwetshiwe futhi okunesikhala esingajwayelekile ngenxa ye-HTML embi. Sakhe le sistimu yokusesha enhle yelabhulali yedijithali enezinkulungwane zamadokhumenti alahliwe. Uhhafu wemibuzo awubuyisanga lutho oluwusizo ngenxa yesikhala esingashintshi. Izisebenzi zasemtatsheni wezincwadi zase zikulungele ukuwuqeda wonke umsebenzi.
Lokhu kuxaka kuba nzima uma usebenza nokuqukethwe okukhiqizwa umsebenzisi, imibhalo ye-OCR'd, noma izilimi ezingasebenzisi izikhala ngendlela efanayo nesiNgisi esenza ngayo (njengesiThai noma isiShayina). Kusho futhi ukuthi lawa mamodeli alwa nama-hashtag, ama-URL, namakhodi omkhiqizo - izinto abantu bangempela abazifunayo nsuku zonke.
Izithenjwa ziyinselele
Amamodeli ashumekiwe abona okuthi "Imoto ingakwesobunxele sesihlahla" kanye nokuthi "Imoto ingakwesokudla kwesihlahla" acishe afane—isilinganiso sokufana esiphezulu esingu-0.98. Ngaphandle kokuchaza imibono ephambene, amamodeli ashumekiwe awabona ecishe afane. Ucabanga ukuthi abaphathi bezindlu ezigcina impahla bazojabula uma amarobhothi eletha amaphakheji eziteshini ezingafanele? Ngingakuqinisekisa ukuthi bengizokhungatheka.
Cabanga ngazo zonke izimo lapho umbono kanye nezinhlaka zereferensi zibalulekile - izikhombisi-ndlela zokuzulazula, ubudlelwano bendawo, ukuma okuhlobene ezinqubweni zezokwelapha, nezincazelo ezingokomthetho zezigcawu zezingozi. Le mehluko ayiyona into encane - ishintsha ngokuphelele incazelo esekelwe ekubukeni kwayo. Empeleni sisebenza namamodeli angakwazi ukusho ukuthi uchaza okuthile okusuka ngaphambili noma ngemuva.
Ama-Counterfactuals Ahlehliswa Ngokuphelele
Lona wangihlekisa, wabe esekhala. Amamodeli ashumeka abona "Uma isidingo sikhuphuka, izintengo zizokhuphuka" futhi "Uma isidingo sikhuphuka, amanani azokwehla" ngokufana ncamashi – amaphuzu ashaqisayo angu-0.95 afanayo . Naphezu kwalokhu, babonisa izimo zomnotho ezihluke ngokuphelele!
Ngifike kule nkinga ngokwakha uhlelo lokuhlaziya amaphepha ocwaningo lwezomnotho. I-algorithm ayikwazanga ukuhlukanisa phakathi kobudlelwano bembangela ephikisayo. Izazi zezomnotho ezazisesha amaphepha amayelana nokukhuphuka kwentengo ngesikhathi sokwanda kwesidingo bathole imiphumela mayelana nokwehla kwentengo ngesikhathi sokwehla komnotho esikhundleni salokho. Ingabe ucabanga ukuthi abahlaziyi bezezimali abenza izinqumo zokutshala izimali zesigidi samadola bangakujabulela ukuthola ulwazi olubuyela emuva? Nginesiqiniseko sokuthi ngeke ngifune ukuthi izimali zami zomhlalaphansi ziphathwe ngaleyo ndlela.
Futhi, cabanga ngazo zonke izimo lapho ukucabanga okungelona iqiniso kubalulekile: ukuqagela kwezomnotho, ubudlelwano bembangela yezokwelapha kanye nomphumela, imibono eqanjiwe yezomthetho, nokuhlaziywa kokwehluleka kobunjiniyela. Uma ungakwazi ukusho ukuthi "uma u-X, bese u-Y" kusukela kokuthi "uma kungenjalo-X, hhayi u-Y", awuqondi kahle ubuhlobo obuyimbangela. Empeleni sisebenza namamodeli angakwazi ukubamba ingqondo enemibandela eyisisekelo naphezu kokuphatha umbhalo ogcwele ukuqagela nokuqagela.
Amabanga Namanani Aqondile
Lona wangishiya nginkemile. Amamodeli okushumeka abona "Umkhiqizo ubiza phakathi kuka-$50-$100" kanye nokuthi "Umkhiqizo ubiza ncamashi u-$101" CISHE into efanayo - amaphuzu afanayo angu-0.98 amangalisayo . Umuntu usebangeni, ungaphandle kwalolo hlu, kodwa imodeli ibona kancane!
Ngithole lokhu ngenkathi ngakha uhlelo lokuqhathanisa amanani eklayenti le-e-commerce. Usesho alukwazanga ukuhlukanisa phakathi kobubanzi bentengo nezintengo eziqondile, ngisho nalapho intengo yayingaphandle kwebanga elishiwo. Abathengi abanesabelomali esiqinile sika-$100 abafuna imikhiqizo "engaphansi kuka-$100" baqhubeka nokuboniswa izinto ezibiza u-$120 noma u-$150. Ingabe ucabanga ukuthi amakhasimende anesabelomali esingaguquki ayakujabulela ukubona imikhiqizo angakwazi ukuyikhokhela ngokusobala? Ngiqinisekile ukuthi ngeke ngifune ukuchitha isikhathi ngiphequlula izinto engingakwazi ukuzithenga.
Cabanga ngazo zonke izimo lapho ububanzi obuqhathaniswa namanani angempela bubalulekile - izinqumo zentengo, ububanzi bemithamo yemithi, izinsuku zokugcina ezisemthethweni, ukubekezelela ukuphepha, nezilinganiso zokusebenza. Uma imodeli yakho iphatha "okungenani izinsuku ezingama-25" kanye "nezinsuku ezingama-20 ncamashi" njengokufana, ulahlekelwe incazelo ebucayi. Empeleni sisebenza namamodeli angakwazi ukuhlukanisa phakathi kokuguquguquka nokunemba naphezu kokuphatha umbhalo; lo mehluko ushayela izinqumo.
Iqiniso Nemiphumela
Nasi isiqhathaniso phakathi kwe -msmarco-distilbert-base-tas-b, all-mpnet-base-v2, kanye ne-open-ai-text-embedding-3-large, futhi uzoqaphela ukuthi awukho umehluko omkhulu phakathi kokuphumayo kwalawa mamodeli.
isikolo sokushumeka se-msmarco-distilbert-base-tas-b kuzo zonke izimo zokuhlola ezihlukene
wonke-mpnet-base-v2 umphumela wokushumeka kuzo zonke izimo zokuhlola ezihlukene
isikolo se-openai-text-embedding-3-enkulu kuzo zonke izimo zokuhlola ezihlukene
Usebenza Kanjani Ngokushumeka
Bheka, ukushumeka kuwusizo ngendlela emangalisayo naphezu kwalezi zinkinga. Angisho ukuthi ungawasebenzisi – ngisho ukuthi uwasebenzise uvule amehlo. Nasi iseluleko sami esivivinywe empini ngemva kwenqwaba yamaphrojekthi nokwehluleka okungenakubalwa:
Hlola imodeli yakho kumaphethini olimi lwangempela lomsebenzisi ngaphambi kokusebenzisa. Akuwona amabhentshimakhi ezemfundo, hhayi amacala okuhlola ahlanzekile - izibonelo zangempela zendlela abasebenzisi bakho abaxhumana ngayo. Sakhe ikhithi yamathuluzi "yokuhlola ingcindezi yolimi" elingisa ukuhluka okuvamile njengokuphikiswa, ama-typos, nokwehluka kwezinombolo. Wonke amasistimu esiwahlolayo ayahluleka kwezinye izindawo - umbuzo ukuthi lezo zindawo zibalulekile yini kuhlelo lwakho lokusebenza oluthile.
Yakha ama- guardrails ezindaweni ezibucayi eziyimpumputhe. Izinhlelo zokusebenza ezahlukene zinezidingo ezihlukile zokungakwazi ukuhluleka. Ngokunakekelwa kwezempilo, ngokuvamile ukuphika kanye nokunemba kwebhizinisi. Ngezezimali, izinombolo kanye nobudlelwano besikhashana. Ngokomthetho, yimibandela nezibopho. Khomba ukuthi yini engeke ihambe kabi esizindeni sakho, futhi usebenzise izivikelo ezikhethekile.
Faka amasu ahlukene esikhundleni sokubheja yonke into eshumekiwe. Amasistimu ethu aphumelela kakhulu ahlanganisa ukubuyisa okusekelwe ekushumekeni nokuqinisekisa kwegama elingukhiye, ukuhlola imithetho ecacile, nezihlukanisi ezikhethekile zokuhlukanisa okubalulekile. Lokhu kuphelelwa umsebenzi akusebenzi kahle; kubalulekile.
Yiba sobala kubasebenzisi mayelana nalokho isistimu engakwazi ukukwenza nalokho engakwazi ukukwenza ngokwethembeka. Sengeze amaphuzu okuzethemba ahlaba umkhosi ngokusobala lapho umphumela ungase uhilele ukunganaki, ukuqhathanisa izinombolo, noma amanye amaphuzu angaba buthaka. Abasebenzisi bayakwazisa ukwethembeka, futhi kwakha ukwethemba uhlelo lulonke.
Nansi into ebaluleke kakhulu engiyifundile: lezi zibonelo aziluqondi ulimi ngendlela abantu abaluqonda ngayo - baqonda amaphethini ezibalo. Lapho ngiyeka ukulindela ukuqonda okufana nomuntu futhi ngiqala ukubaphatha njengamathuluzi ayinkimbinkimbi okufanisa iphethini anezindawo ezithile eziyimpumputhe, amasistimu ami aba ngcono. Okungcono kakhulu.
Izindawo eziyimpumputhe engizichazile ngeke ziphele maduzane - sezibhekwe ukuthi lezi zinhlobo zisebenza kanjani. Kodwa uma wazi ukuthi akhona, ungaklama okuzungezile. Futhi ngezinye izikhathi, ukuvuma ukulinganiselwa kuyisinyathelo sokuqala sokukunqoba.
Qaphela: Nginezimo eziningi ezinjalo ezitholakala ngokuhlolwa, futhi ngizozimboza kokuthunyelwe kwami okulandelayo kanye nezibonelo zekhodi.
Isihloko esilandelayo esiqhubekayo sizophuma maduze. Hlala ubukele!!
