Ibambekile Esikhathini: Kungani i-AI Ingakwazi Ukuyeka Ukudweba Amawashi ngo-10:10

Masikhulume iqiniso, i-AI ipholile ngendlela emangalisayo—kuze kube yilapho ibikezelwa ngendlela emangalisayo.

Njengamanje, kungenzeka ukuthi usuzibonile izibonelo zokweba izihloko zezindaba ze-AI ekhiqizayo ehlanganisa ubuciko be-surreal, okubonakalayo okukhazimulayo, noma imiklamo engenakwenzeka yokudala. Icele ukuthi icabange amadolobha angaphandle egcwele ukukhanya kwe-neon noma amahlathi lapho izihlahla zimila khona izimbali ze-bioluminescent, futhi—i-boom!—uvezwa isithombe esidlula imingcele yalokho abantu abavame ukukucabanga.

Kodwa-ke, ucela i-AI ukuthi idwebe iwashi. Futhi yonke imilingo iyama. Utholani? Iwashi lanamathela ngo-10:10 ngenkani.

Kucishe kuhlekeke: noma ngabe uyikhuthaza kanjani i-AI—“dweba iwashi lesandla lakudala!” “iwashi lekusasa!” noma ngisho “newashi elincibilikile elifana ne-Dali!”—lezo zandla zewashi ngandlela thize zithola indlela yazo lapho zifika lapho okumangalisayo okuwu-10:10. Uma i-AI kufanele iqonde i-nuance, randomness, kanye nobuciko, kungani ibambelele kangaka kulokhu?

Impendulo ayiyona nje i-artifact ehlekisayo yamamodeli okuqeqesha kodwa ukubuka okuncane kwezinselelo ezinkulu i-AI ebhekana nazo uma kuziwa ekuqondeni ubuhlakani, ukwenzelela, kanye nokugqashula ezingqungqutheleni ezigqokwe kahle. Ngakho-ke, qinisa ibhande lakho esihlakaleni, futhi ake sijule kule mfihlakalo yefilosofi emangalisayo—nobuchwepheshe obujulile.

Isenzakalo 10:10: Ifa Lomuntu

Ngaphambi kokuthi siqale ukunyakazisa iminwe ku-AI, ake sikhulume ngathi . Isizathu sokuqagela kwe-AI ku-10:10 asiveli ku-algorithm yokunquma, "Yebo, kulapho isikhathi sizizwa siphelele khona." Cha—imane ivuselela impatho thina bantu esiye sayenza edizayini yewashi amashumi eminyaka.

Cishe sonke isikhangiso sewashi owake wasibona sisebenzisa isitembu sesikhathi esifana nesika-10:10. Futhi cha, lokhu akubangelwa ukuthi wonke umthwebuli wezithombe womkhiqizo emhlabeni wonke ujoyine ihlelo le-"10:10." Nakhu ukuthi kungani le nketho yesikhathi ibusa kakhulu:

1. I-Symmetry Ibukeka Yinhle : Ngo-10:10, izandla zewashi zenza umuzwa omuhle wokuvumelana okubonakalayo. I-symmetrical, kodwa ayiqinile kakhulu. Iphinde ifake ilogo yomkhiqizo kahle, evamise ukushayeka endaweni yehora le-12 emawashini amaningi.

   
   

2. Umthelela we-'Smiling Watch' : Bhekisisa kahle: Ngo-10:10, izandla ezigobele phezulu zilingisa isimo sokumomotheka. Ngokuqaphela noma ngokuzithoba, amabhrendi ayaqonda ukuthi izinkomba zedizayini ezijabulisayo, ezamukelayo zithengisa imikhiqizo eminingi.

   
   

3. Ukugcwala Kwezokukhangisa : Uma le ngqungquthela isiba namandla, lakhithika. Kusukela ezikhangisweni kuya ezithombeni zesitoko kuya ezithombeni zekhathalogi, yonke indawo iwashi lalivela, 10:10 kwakuyindinganiso. Kwaba umthetho wokuklama oziqhubayo.

Sekungamashumi eminyaka sondla umhlaba lokhu okubonakalayo, sikwenza kutholakale yonke indawo kangangoba ngisho nengqondo yethu izenzakalela kukho uma sicabanga ubuso bewashi. Asicabangi nokucabanga ngakho—sivele sikulindele.

Futhi manje, i-AI nayo iyakwenza.

Inkinga yesibuko se-AI

Ukuze uqonde ukuthi kungani i-AI, ngezinye izikhathi ebizwa ngokuthi “umlingisi omkhulu,” ingakwazi ukugqashula ku-10:10, ake sikhiphe ngokushesha ukuthi lezi zinhlobo zifunda kanjani.

Yonke imodeli ye-AI ekhiqizayo—kuhlanganise nezindlu zamandla ezifana ne-Stable Diffusion, i-DALL-E 2, ne-MidJourney—ithembele kumasethi edatha amakhulu ukuze iqeqeshwe. Lawa maqoqo edatha amaqoqo amakhulu ezithombe (ngokuvamile izigidigidi) ezikhishwe ku-inthanethi: ukuthwebula izithombe zesitoko, amakhosombe aku-inthanethi, okuqukethwe okukhiqizwa umsebenzisi, ukusho lokho.

Uma i-AI ifunda umqondo “wewashi” kulezi zithombe, ayigcini nje ngokuhlaziya ubuhle noma umsebenzi wewashi. Ibheka amaphethini okuphindaphinda.

Qagela ukuthi yini ebusa izithombe ze-inthanethi zamawashi? Yebo, 10:10.

"Engqondweni" ye-AI engagxeki, iqiniso elibaluleke kakhulu ngokwezibalo mayelana namawashi akukhona ukuthi asho isikhathi. Ukuthi cishe njalo zibukeka kanjena:

• Izandla ezilinganayo zikhomba ku-10 no-2.
• I-logo ehlezi ngokuziqhenya kumaki wehora le-12.
• Futhi, ngezinye izikhathi, izinkinga zebhonasi ezifana nokudayela kwe-chronograph kubekwe njengokugqoka kwewindi.

Uma u-95% wezithombe "zewashi" i-algorithm ebonwayo zifana ncamashi, qagela ukuthi kwenzekani uma uyicela ukuthi idale iwashi? I-AI ayazi kangcono. Icabanga ukuthi ufuna noma yiluphi uhlobo lwewashi olujwayeleke kakhulu kulo—10:10.

Kodwa Linda—I-AI Ayigcini Ngokulandela Idatha... Kulungile?

Ungase ucabange: “Yima kancane, i-AI kufanele ibe yokudala! Kungani ingahlubuki?”

Yilapho izinto ziba nzima khona. I-AI ingase ibonakale inobuhlakani—njengokungathi ikhipha imibono emoyeni omncane—kodwa akunjalo. Esikhundleni salokho, isebenza ngokungenzeka, idonsa emaphethini efundwe ngesikhathi sokuqeqeshwa. Ake ngikususe lokho.

Cabanga ngobuchopho be-AI njengomdlalo omkhulu “wokuqedela ngokuzenzakalela.” Cabanga uthayipha "izinhlobo zezinja" ku-Google—iziphakamiso zokuqedela ngokuzenzakalela ezifana ne-"Labrador" noma i-"German Shepherd" ziyavela ngoba zivame kakhulu. Ngokufanayo, uma i-AI ikhiqiza isithombe "sewashi lesandla," isampula ukuthi icabanga ukuthi iwashi elijwayelekile libukeka kanjani ngokusekelwe kumaphethini eselivele liwabonile.

Nansi imininingwane ebalulekile yobuchwepheshe:

Amamodeli akhiqizayo adala izithombe ngokuhlola "indawo ecashile," ukumelwa kwezibalo okuphezulu kwakho konke abakufundile. Ake ucabange lesi sikhala esicashile njengomthala ominyene owenziwe ngamaphethini, imibono, nokuma. Izinto ezifana “nobuso bewashi” zakha amaqoqo kulo mthala, futhi endabeni yamawashi... ingxenye eminyene kakhulu, efinyeleleka kalula yalelo qoqo ngu—ukuqagele—10:10.

Lapho imodeli iqala ukukhiqiza isithombe, lezi zindawo eziminyene zisebenza njengemithombo yamandla adonsela phansi. Kungenzeka kakhulu ukuthi ukhethe okuthile okuseduze kunokuba uphambukele "ekungenzeki kokudala."

Ukugoqa Kwemodi: I-Trap AI Ayikwazi Ukuphunyuka

Kukhona futhi okunye okudlalwayo lapha: ukugoqa kwemodi.

Ukugoqa kwemodi kuwumgodi ovamile ekufundeni komshini lapho imodeli ye-AI iqala ukuthanda ingxenye encane yamathuba, ingaziba izinketho ezibonwa kancane. Kufana nesibani esikhanyisa izibonelo ezivame kakhulu kuyilapho ezinye zifiphala zibe mnyama. Ngoba amawashi ngo-10:10 amelwe ngokweqile kumadathasethi okuqeqeshwa kwe-AI, aphenduka “okuzenzakalelayo.” Ngaso sonke isikhathi uma utshela i-AI, ibuyela emuva kulokhu kukhetha okuphephile nokujwayelekile.

Nansi into: lokhu akugcini nje ngamawashi. Ukuchema okufanayo kungena kuzo zonke izinhlobo zezinto ezikhiqizayo. Cela i-AI ukuthi yenze, ake sithi, isithombe esijwayelekile “sosomabhizinisi,” futhi ngokuvamile uzothola owesilisa waseNtshonalanga ogqoke isudi nothayi—ngoba yilokho okubusa izithombe zesitoko. I-AI ayichemile njengedatha yayo—futhi amasethi edatha, njengoba sazi, agcwele amashumi eminyaka, ngisho namakhulu eminyaka, okuchema kwabantu.

Linda... Asikwazi Nje Ukukulungisa?

Ngokwethiyori, yebo. Ngobuchwepheshe? Kuyinadi eliqine kakhulu ukuliqhekeza.

Ukuze i-AI iphume endleleni yayo engu-10:10—noma okunye ukuchema okugxilile kwamasiko—idinga idatha nama -algorithms amelana ngokugcwele nenetha yokuphepha yesilinganiso. Nakhu ukuthi lokho kungase kubukeke kanjani:

1. Diversifying Datasets : Okokuqala, qinisekisa ukuthi amasethi edatha okuqeqesha afaka ezinye izindlela ezingameli kahle . Uma idatha yokuqeqeshwa kwe-AI ifaka amawashi ngezikhathi ezingahleliwe kaningi ngo-10:10, singathambisa lokhu kuchema. Kodwa ukukala lokhu kumadathasethi amakhulu akuyona into encane—futhi ukuhlanza amasethi edatha kuthatha inhlanganisela ebalulekile yezinsizakusebenza zabantu.

   
   

2. Amathuba Okwenza Kabusha : Onjiniyela bangakwazi ukulungisa ama-algorithms omvuzo we-AI ukuze bakhuthaze ngenkuthalo imiphumela engajwayelekile. Isibonelo, bangangeza izinhlawulo zokudonsa ngamandla kakhulu emiphumeleni ezenzakalelayo njengo-10:10.

   
   

3. Ukujova Umsindo Ezazisweni : Amasistimu athuthukisiwe angethula “umsindo osheshayo,” ngokusobala aphoqelele i-AI ukuthi ihlele ngokungahleliwe izici ezicashile zemiphumela yayo, njengokuma kwezandla ewashini—noma, kabanzi, ukuhlola izindawo ezingahloliwe kahle zesikhala esicashile.

   
   

4. Ukushuna Kahle Ngokwezifiso : Amamodeli angaphinda ashunwe kahle ukuze agudluze okudaliwe kuye ekudaleni okukhulu. Ngokuqeqesha amamodeli amancane, akhethekile kudatha ehlukahlukene kakhulu noma ye-niche (njengedathasethi yamawashi ngo-7:13 noma 4:47), abadali bangase bacheme imiphumela ethile ekwephuleni isikhunta.

Sekushiwo, kunomthambeka oshelelayo lapha. Ukukhuthaza ukungahleleki okuningi kusho ukuthi i-AI ingalahlekelwa ukusekela kwayo ngokuphelele, idale imiphumela ezizwa ingahlangani noma ingenangqondo kunokuthi “iwukudala.” Ukuthola indawo emnandi phakathi kwamaphethini azenzakalelayo kanye nokuqamba okusha kweqiniso kuhlala kungenye yezinkinga ezinkulu ekuthuthukisweni kwe-AI namuhla.

Ngakho-ke, Iyini I-Takeaway Enkulu?

Isizathu sokuthi i-AI igcine ukudweba amawashi ebambelele ku-10:10 akukhona nje mayelana nedatha yayo yokuqeqeshwa noma izici zokubhala amakhodi—i-microcosm yokuthi i-AI ekhiqizayo ibonisa kanjani imikhawulo yokusungula kwethu, ukuchema kwethu, kanye nedatha yethu. Uma silindele ukuthi i-AI "icabange ngaphandle kwebhokisi," siyakhohlwa ukuthi yakhiwe ngaphakathi kwebhokisi lethu ukuqala.

Okungihlaba umxhwele ngalokhu akuyona i-humdrum yobuchwepheshe yokuthi izikhala ezifihlekile noma ukusatshalaliswa kokuqeqeshwa kusebenza kanjani (yize ngizovuma, lokho kuhle kakhulu ngokwako). Okugqamayo lapha ukuthi i-AI isiphoqa kanjani ukuthi sicabange ngamaphethini ethu . Senze u-10:10 waba uphawu lwendawo yonke lwamawashi. Futhi kuze kube yilapho sishintsha imihlangano yethu—noma sifundise i-AI ukuthi yazise ukuhlukahluka ngaphezu kokujwayelana—izoqhubeka nokunanela lezo zinqumo zibuyela kithi.

Ngakho-ke, ngokuzayo uma ucela i-AI ukuthi idale iwashi elibambelele esikhathini esedlule, kuthathe njengesikhumbuzo esimnene: ubuhlakani abuhlali bumayelana nama-algorithms. Kumayelana nenhloso.

Futhi okwamanje, ubuso bewashi le-AI busamamatheka kuwe, buqandiswe ihora le-10 lidlule ihora leshumi.