Izisekelo Zobuhlakani Bokwenziwa - Ingxenye Yesithathu: Imiqondo
Yebo Uzithandile yini izikhala zethu zokuqala ezimbili ezihlobene nokufunda kwe-Artificial Intelligence (AI) kubasebenzisi be-Linux?Ngakho-ke namuhla cishe uzoyijabulela le ngxenye yesithathu, lapho sizoxoxa khona ngemibono eminingana ebaluleke kakhulu kanye namagama ahlobene nale nsimu yezobuchwepheshe. Futhi uma ungakafundi izingxenye ezimbili ezedlule, singathanda ukucacisa lokho. Engxenyeni yesibili sizokhuluma iphuzu mayelana nomehluko phakathi Ubuhlakani Bokwenziwa Obungokomfanekiso kanye Nobuhlakani Bokwenziwa Bokuxhumana kanye nezigaba zomlando zalo mkhakha wezobuchwepheshe eminyakeni eyikhulu edlule. Kanti, ku- Engxenyeni yokuqala sigxila en Kuyini i-AI futhi ngezinhlobo ezahlukahlukene esezivele zenziwe (Encane noma Ebuthakathaka, Ekhiqizayo kanye Ne-Agent) futhi okulindeleke ukuthi kudalwe (Okujwayelekile noma Okuqinile, kanye Nobunhloli Bokwenziwa).
Futhi njengasezimweni zangaphambilini, khumbula ukuthi njengasezinsukwini zabo ugesi, umsakazo, ithelevishini, amakhompyutha, izingcingo, i-inthanethi kanye nezinkundla zokuxhumana; Ubuhlakani Bokwenziwa buwubuchwepheshe obufinyelele ekuhlanganisweni ngokujulile kukho konkeFuthi lokho, ngokusongela isikhathi esibalulekile namahora okusebenza abizayo, nokusenza sikhiqize kakhudlwana, sibe nobuciko, futhi sikwazi ukushintshashintsha; ngokuqinisekile, Kancane kancane, izosusa labo abangenalo nombono wokuthi ikhona futhi isebenza kanjaniNgakho-ke, akukho okungcono kunokuqala ukuthola umongo ngakho konke okuhlobene nakho.
Izisekelo Zobuhlakani Bokwenziwa - Ingxenye Yokuqala: Imvelaphi II
Kodwa, ngaphambi kokuba singene kulokhu "Ingxenye yesithathu yensimu ethokozisayo nemangazayo yoBuhlakaniphi Bokwenziwa"Futhi ngokuqondene neminye yemiqondo namagama abaluleke kakhulu okufanele siwacabangele, siwaqonde, futhi siwaqonde kahle, sincoma ukuhlola incwadi yangaphambilini neyamuva ehlobene kulolu chungechunge lwezincwadi, lapho eqeda ukuyifunda:
Izisekelo Zobuhlakani Bokwenziwa – Ingxenye Yesithathu: Imiqondo Nemigomo Ehlobene
Imiqondo namagama abalulekile angu-20 ahlala ekhulunywa ngawo emkhakheni we-Artificial Intelligence
1. Ukuphupha izinto ezingekho
Ukuphupha izinto ezingekho emthethweni kubhekisela esimweni lapho uhlelo lwe-AI, ikakhulukazi oluphatha ukucubungula ulimi lwemvelo, luqala ukukhiqiza imiphumela engase ingabalulekile, engenangqondo, noma engalungile ngokusekelwe kokufakwayo okunikeziwe. Lokhu kuvame ukwenzeka lapho uhlelo lwe-AI lungaqiniseki ngomongo, luthembele kakhulu kudatha yalo yokuqeqesha, noma lungenalo ukuqonda okufanele ngendaba.
2. Ukufunda ngomshini
Insimu encane ye-AI ehlanganisa ukuthuthukiswa kwama-algorithms namamodeli ezibalo avumela imishini ukuthi ithuthukise ukusebenza kwayo ngolwazi. Isibonelo kungaba i-algorithm yokufunda komshini engabikezela amathuba okuthi amakhasimende angaphumeleli ngokusekelwe ekuziphatheni kwesikhathi esidlule. Ngenxa yohlelo lokufunda komshini, umshini ufunda ukuxazulula izinkinga ngezibonelo: ngaleyo ndlela ungaqhathanisa futhi uhlukanise idatha futhi uqaphe ngisho nokwakheka okuyinkimbinkimbi.
3. Ukufunda Okujulile
Insimu encane yokufunda komshini esebenzisa amanethiwekhi e-neural anezingqimba eziningi ukuze ifunde kudatha, okusho ukuthi isebenzisa izakhiwo ezahlukene zenethiwekhi ye-neural ukuze ifunde izendlalelo ezilandelanayo zokumelwa kwedatha okunenjongo ngokwengeziwe. Njengamanje, iyinsimu ethandwa kakhulu ngaphakathi kwe-AI. Yavela ekuhlanganisweni kwama-algorithms okufunda komshini namanethiwekhi e-neural asemthethweni kanye nokusetshenziswa kwe-big data.
4. Idatha Enkulu
I-Big Data iyigatsha le-Information Technology elifunda ubunzima obukhona bokuphatha amasethi amakhulu edatha. Elibizwa nangokuthi i-Macrodata, lawa abhekisela kumasethi amakhulu nayinkimbinkimbi avame ukuba makhulu kakhulu noma ahlukahlukene ukuthi angacutshungulwa kusetshenziswa izindlela zendabuko. Ngaphezu kwalokho, avela emithonjeni ehlukahlukene (amanethiwekhi omphakathi, izinzwa, njll.) futhi adinga amasu athuthukile, njengobuhlakani bokwenziwa kanye nokufunda komshini, ukuze ahlaziye futhi akhiphe ulwazi oluwusizo.
5. Isethi yedatha
Isethi yedatha iqoqo ledatha, ngokuvamile elakhiwe etafuleni. Liyiqoqo ledatha elihleliwe elisetshenziselwa ukuqeqesha, ukuhlola, noma ukuqinisekisa amamodeli obuhlakani bokwenziwa. Lingaqukatha umbhalo, izithombe, amavidiyo, noma ezinye izinhlobo zolwazi, ngokuvamile ezilebula ukuvumela ama-algorithms okufunda komshini ukuthi aqaphele amaphethini futhi enze izibikezelo.
6. Ukuqeqeshwa kwemodeli
Ukuqeqeshwa kwemodeli kuyinqubo yokusebenzisa isethi yedatha ukufundisa ubuhlakani bokwenziwa noma imodeli yokufunda komshini ukwenza umsebenzi othize, njengokuhlukanisa noma ukubikezela. Imodeli ilungisa amapharamitha ayo ngokusekelwe ezibonelweni ezinikeziwe ukuze ithuthukise ukunemba kwayo ngedatha entsha.
7. Ukuthuthukiswa Kokukhiqizwa Ngokululama
I-RAG (Ukuthola Okungeziwe Kokukhiqizwa) iyindlela evumela i-AI ukuthi ibuze izizindalwazi zangaphandle noma iseshe i-inthanethi ukuthola ulwazi olusha nolubuyekeziwe ngaphambi kokukhiqiza impendulo. Esikhundleni sokudinga ukuqeqesha kabusha yonke imodeli ngedatha entsha, i-RAG isebenza njengohlobo "lwemfundo yesikhathi sangempela" ye-AI. Lokhu kuwusizo kakhulu ngoba kuphoqa uhlelo ukuthi luthembele kumadokhumenti athile futhi aqinisekisiwe, okunciphisa kakhulu ingozi yokuthi i-AI iqambe ulwazi (ukuphupha).
8. I-GPT (I-Generator Pre-trained Transformer)
Bangumndeni wamamodeli olimi athuthukiswe yi-OpenAI asebenzisa amanethiwekhi e-neural ukuqonda nokukhiqiza umbhalo ngokwemvelo. Lawa mamodeli e-GPT (Generative Pretrained Transformer) asebenzisa isakhiwo esibizwa ngokuthi i-Transformer esibavumela ukuthi bahlaziye izimpendulo futhi babikezele impendulo engaba khona ngokusekelwe ekuqeqeshweni ngamasethi edatha amakhulu. Ngenxa yalokhu, i-GPT iyakwazi ukulingisa amakhono okuqonda ulimi lwabantu.
9. Isizukulwane Solimi Lwemvelo (Isizukulwane Solimi Lwemvelo)
Ukwenziwa kolimi lwemvelo kuyinkambu encane yobuhlakani bokwenziwa ehilela ukukhiqiza ngokuzenzakalelayo imibhalo noma inkulumo eqondakalayo olimini lwabantu. Ivumela umshini ukuguqula idatha eluhlaza ibe imisho noma izigaba zemvelo, njengezifinyezo ezizenzakalelayo noma abasizi abangokoqobo.
10. Ubunjiniyela Obusheshayo
Kungubuciko bokuthuthukisa izikhuthazo zezicelo zokufeza imiphumela ekhiqizayo enenjongo—okungukuthi, imiphumela enembile, ekhiqizayo, nemisha. Noma, ngenye indlela, kuyisayensi yokuklama, ukwenza ngcono, kanye nokucwengisisa imiyalelo esiyinika ubuhlakani bokwenziwa obukhiqizayo. Umgomo wayo ukuqondisa i-AI ukuthola izimpendulo ezinembile, ezifanele, nezisezingeni eliphezulu, esebenza njengebhuloho lokuxhumana ukuze kugwenywe imiphumela ejwayelekile noma eyiphutha.
11. Imodeli
Imodeli yinto ezomela umphumela we-algorithm yokufunda. Ngakho-ke, yilokho i-AI ekusebenzisayo ukwenza izibikezelo zomphumela. Isibonelo, amamodeli ayisisekelo ayisigaba esibanzi samamodeli e-AI afaka amamodeli olimi olukhulu kanye nezinye izinhlobo zamamodeli afana namamodeli okufunda ombono wekhompyutha kanye nokuqinisa. Abizwa ngokuthi "amamodeli ayisisekelo" ngoba asebenza njengesisekelo sokwakha izinhlelo zokusebenza, ebhekana nohlu olubanzi lwezizinda kanye namacala okusetshenziswa.
12. Amamodeli Olimi Amakhulu
Ziwuhlobo lwemodeli yokufunda okujulile eqeqeshwe kusethi enkulu yedatha ukuze yenze imisebenzi yokuqonda ulimi lwemvelo kanye nokukhiqiza. Amamodeli adumile afaka i-BERT, i-PaLM, i-GPT-2, i-GPT-3, i-GPT-3.5, kanye ne-GPT-4 entsha. Lawa mamodeli ayahlukahluka ngobukhulu (inani lamapharamitha angalungiswa), ububanzi bemisebenzi angayenza (ukubhala ikhodi, ukuxoxa, isayensi, njll.), kanye nesisekelo solwazi aqeqeshwe kuso.
13. Amamodeli Olimi Oluncane
Ziyizinhlobo ezincane kakhulu zamamodeli ezilimi ezinkulu, noma ama-LLM. Zombili zisebenzisa amasu okufunda komshini ukuze zibone amaphethini nobudlelwano ukuze zikwazi ukukhiqiza izimpendulo zolimi lwemvelo ezingokoqobo. Kodwa ngenkathi ama-LLM amakhulu futhi edinga amandla amaningi okusebenzisa ikhompyutha kanye nenkumbulo, amamodeli ezilimi amancane afana ne-Phi-3 aqeqeshwa kumasethi edatha amancane, ahlelwe kakhulu futhi anamapharamitha ambalwa, okwenza abe mancane futhi asebenziseke ngisho ungaxhunyiwe ku-inthanethi.
14. Ubandlululo
Lokhu kuyizinto eziqagelwayo ezenziwe yimodeli ye-AI mayelana nedatha. "Ukuhwebelana ngokubandlulula" yibhalansi okumele ifinyelelwe phakathi kwezinto eziqagelwayo ezenziwa yimodeli mayelana nedatha kanye nenani izibikezelo zemodeli ezishintshayo, uma kunikezwe idatha yokuqeqeshwa ehlukile. Ukubandlulula okwenziwayo yisethi yeziqagelwayo ezenziwa yi-algorithm yokufunda komshini mayelana nokusatshalaliswa kwedatha okuyisisekelo.
15. Ukucutshungulwa Kolimi Lwemvelo
Kuyinsimu ehlanganisa imikhakha eyahlukene, umqondo wayo oyinhloko ukunikeza imishini ikhono lokufunda nokuqonda izilimi abantu abazikhulumayo. Ucwaningo lwe-Natural Language Processing luhlose ukuphendula umbuzo wokuthi abantu bakwazi kanjani ukuqonda incazelo yomusho okhulunywayo noma obhaliwe nokuthi baqonda kanjani okwenzekile, nini futhi kuphi; kanye nomehluko phakathi kokucabanga, inkolelo, kanye neqiniso. Ngenxa yalokho, kuhilela ukuhlela amakhompyutha ukuze kucutshungulwe inani elikhulu ledatha yezilimi futhi kugxile ekuguquleni umbhalo okhululekile ube isakhiwo esijwayelekile.
16. Isaziso
I-prompt iyimiyalelo noma isethi yemiyalelo (okufakwayo) enikezwa uhlelo lwekhompyutha ukuze ikhiqize impendulo. Ngokomongo we-Artificial Intelligence, iyimiyalelo yokuqala, umyalo, umbuzo, isicelo, noma isitatimende sombhalo esinikezwa imodeli ukuze siqondise futhi siqondise ukukhiqizwa kwempendulo noma umphumela othize. I-prompt isebenza njengesikhuthazo noma isisusa se-AI, esibonisa isihloko, isitayela, isiqondiso, noma uhlobo lwempendulo olulindelekile. Indlela i-prompt eyakhiwe ngayo (ukwakheka kwayo) inquma ngqo ikhwalithi nokufaneleka kwemiphumela etholiwe.
17. Inethiwekhi Yezinzwa
Ziyimodeli yokubala esekelwe kusethi enkulu yamayunithi e-neural alula (ama-neuron okwenziwa), afana kakhulu nokuziphatha okubonwa kuma-axon ama-neurons ebuchosheni bezinto eziphilayo. Ziyiyunithi yokusebenza eyisisekelo ye-Deep Learning. Ngaphezu kwalokho, iyimodeli yokufunda komshini evumela ukuqashelwa kwamaphethini nokuxazulula izinkinga, ngokusekelwe ekufundeni komshini noma ekufundeni okujulile, efuna ukulingisa izindlela zokuhlaziya zobuchopho bomuntu.
18. I-Transformer
Ama-Transformers amelela uhlobo lokwakhiwa kwenethiwekhi ye-neural eyenzelwe ukucubungula idatha elandelanayo, njengombhalo. Ama-Transformers afunda umongo, futhi ngenxa yalokho, incazelo. Abamba ubudlelwano phakathi kwezinto ezahlukene zokulandelana (amagama, imisho) ngokufanayo, kunokuba alandelelane njengamamodeli endabuko. Ama-Transformers ayisisekelo samamodeli aphumelelayo njenge-GPT ne-BERT futhi asetshenziselwa imisebenzi yokucubungula ulimi lwemvelo.
19. Ithokheni
Ithokheni ku-Artificial Intelligence iyiyunithi eyisisekelo yedatha (amagama, izingxenye zamagama, izinhlamvu, noma izimpawu) ezisetshenziswa amaModeli Okufunda Ulimi (ama-LLM) afana ne-GPT noma i-Gemini ukucubungula, ukuqonda, nokukhiqiza umbhalo. "Imali" noma ulwazi olubalulekile olufundwa futhi luhlaziywe yi-AI, oluvumela ukuthi luhlukanise ulimi lwabantu lube yizicucu ezilawulekayo (ezibalekayo). Ngenxa yalokho, amamodeli e-AI acubungula amathokheni ukuze aqonde ubudlelwano phakathi kwawo futhi avule amakhono afana nokubikezela, ukukhiqiza, kanye nokucabanga. Lapho amathokheni ecubungulwa ngokushesha, amamodeli angafunda futhi aphendule ngokushesha.
20. Ukuqinisekiswa okuphambene
Kuyindlela esetshenziswa ukuhlola imiphumela yokuhlaziywa nokuqinisekisa ukunemba kwezibikezelo zemodeli yokufunda komshini. Ngakho-ke, kuyindlela yokuhlola amamodeli okufunda komshini egxile ekuhlukaniseni isethi yedatha ibe yizingxenye eziningana (noma "ama-fold"). Imodeli bese iqeqeshwa kwezinye izingxenye bese ihlolwa kwezinye. Lokhu kuvumela ukulinganisa okuthembekile kokusebenza kwemodeli ngokunciphisa ukufunda ngokweqile.
Ezinye izisekelo zoBuhlakaniphi Bokwenziwa okufanele ziphenywe, ziqondwe futhi zifundwe kahle kule ngxenye yesithathu
Lezi yizimbalwa nje zemibono namagama abalulekile kwawaningi okufanele ngempela ukuwazi, ukuwaqonda, nokuwaqonda. kule nsimu yesimanje nentsha yoBuhlakaniphi Bokwenziwa. Ngakho-ke, ukuze sandise ulwazi olunikezwa lapha futhi sikuqondise kalula kwabanye, sikunikeza lokhu I-infographic ethakazelisayo eyenziwe nge-AI ukukunikeza umbono omncane ngabanye ababaluleke ngokulinganayo nababalulekile:
Isifingqo
Ngamafuphi, sithemba ukuthi lokhu "Ingxenye yesithathu yensimu ethokozisayo nemangazayo yoBuhlakaniphi Bokwenziwa" Ngokugcizelela i-Generative Artificial Intelligence (GAI), kungenzeka ukuthi inikele kakhulu ekuqiniseni nasekukhuliseni ulwazi lwakho namakhono kulo mkhakha wesimanje nowokusungula izinto ezintsha. Ungakhohlwa ukuthi, Ubuhlakani bokwenziwa kanye namarobhothi kushintsha umhlaba kanye nendlela esiphila ngayo izimpilo zethu.Futhi eminyakeni ezayo, lapho i-quantum computing isitholakala kwezentengiselwano futhi amandla e-quantum (akhiqizwa ngomphumela we-Casimir) etholakala kabanzi kubo bonke abantu, kungaba okuhlukile ngokuphelele endleleni esibhekana ngayo nawo namuhla. Ngakho-ke, kungcono ukufunda nge-Artificial Intelligence manje kunokwenqaba nokuzisola kusasa. Futhi-ke, ungakhohlwa ukuqhubeka nokubamba iqhaza, ukufaka isandla, nokusekela i-Linuxverse yethu esiyithandayo neyigugu.
Okokugcina, khumbula vakashela wethu «ikhasi lasekhaya» ngeSpanishiNoma, nganoma yiluphi olunye ulimi (ngokwengeza izinhlamvu ezimbili kuphela ekugcineni kwe-URL yethu yamanje, isibonelo: ar, de, en, fr, ja, pt kanye no-ru, phakathi kwezinye eziningi) ukuze uthole okunye okuqukethwe kwamanje.