Ukujongana nemiba entsonkothileyo ebandakanya amalungelo abantu kunye neenkcukacha zobuqu kufuna ingqalelo, ubuchule kunye nengxoxo eyimfuneko phakathi kweengqondi neengcaphephe zexesha lethu. Sifumanisa ukuba asikhawulezi ngokwaneleyo ukulungelelanisa imithetho yentlalo kwimingeni eyenziwa yintsha yezobuchwepheshe kuthi. Ubuchwephesha obusakhulayo bufumana ngakumbi ukuba busebenza kwindawo evulekileyo, ngokungabikho ngokupheleleyo kwemimiselo ethintela isicelo sabo, ngokukhululekile ukwenza umonakalo kwaye ngoko ke ukwenza oko ngokungenasiphelo.
Ngaba kunokwenzeka ukucinga ulawulo olubuyela emva kwikhonkco lophuhliso lwezobuchwepheshe kuphando lwezenzululwazi kunye neenjongo zalo zobuchule?
Ngaba kuyacingeleka ukulawula ukuvela kwezinto eziphilayo ngelixa sigcina intlonipho eqinileyo yenkululeko yomntu ngamnye?
Okukhona uzama ukuzifihla, kokukhona utsala umdla. Kutheni le nto kubaluleke kangaka ukuba kungabikho mntu waziyo ngawe?” - ukusuka kwifilimu ethi "Anon" ebhalwe kwaye ilawulwa ngu-Andrew Niccol - 2018
Kwimuvi "Anon” ye-2018, uluntu lwexesha elizayo yindawo emnyama, phantsi kolawulo oluthe ngqo lwenkqubo yekhompyutha enkulu ebizwa ngokuba yi-Ether, ekwazi ukubeka iliso kuzo zonke iikona zesizwe ngokuyiqwalasela ngamehlo abantu abafanayo abahlala kuyo. Wonke umntu ungumveleli egameni likaEtere kwaye uxanduva lwabo lokuqala, ngokuqinisekileyo, kukuzijonga ngokwabo kunye nokuziphatha kwabo.
U-Ether ngowona mhlobo ubalaseleyo wemikhosi yamapolisa: nge-Ether, ii-arhente ziyakwazi ukulandelela amava akhe nawuphi na umntu ngokuzihlaziya ngamehlo abo kunye nokusombulula naluphi na uhlobo lolwaphulo-mthetho.
Ipolisa uSal liyazibuza ukuba kutheni kufuneka ulwe ukukhusela ubumfihlo bakho: yintoni inqaku xa ungenasizathu sokuzifihla? Ngapha koko, kwixesha apho itekhnoloji esakhayo ukwandisa ukhuseleko lwamakhaya ethu kunye nezitrato zethu zifuna ukurekhodwa, ukubeka iliso kunye nokuqinisekiswa kolwazi olunjalo ngokusemdleni wabantu abacela ukhuseleko, sinokulindela njani ukuqinisekisa ubumfihlo babo?
Ukubonisa ukuba kuyingozi kangakanani ukufikelela kubomi babanye, i-hacker iya kuthatha ulawulo luka-Ether kwaye i-nightmare eyoyikisayo iya kuhla kubomi bezigidi zabantu: isoyikiso sokubukela njengababukeli abangenakuzinceda imifanekiso yezona zinto zihamba phambili. Amaxesha abuhlungu obomi babo, asasazwa ngokuthe ngqo kwiiretina zabo.
Le iinethiwekhi ze-neural ezenziweyo esisiseko sokusebenza kobukrelekrele banamhlanje bokwenziwa, ijikeleza kwizinto ezintathu eziphambili: ulwazi olusisiseko olubizwa ngenye indlela. Corpus, un i-algorithm ukudibanisa ulwazi kunye a imemori ukuze bakhumbule.
I-algorithm ayiphelelanga ekulayishweni kwe-banal kwimemori, iyayihlola ekukhangeleni izinto ezihambelana nazo. Umxube wedatha kunye nobudlelwane ziya kudluliselwa kwimemori eya kwenza a itemplate.
Ngaphakathi kwemodeli, idatha kunye nobudlelwane azibonakali ngokupheleleyo, yiyo loo nto ukwakhiwa kwakhona kwe-corpus yolwazi loqeqesho lwangaphambili olusuka kuthungelwano lwe-neural oluqeqeshiweyo phantse akunakwenzeka.
Oku kuyinyani ngakumbi xa ii-corpus ziqulethe inani elikhulu ledatha. Le yimeko yeenkqubo ezinkulu zeelwimi ezaziwa ngokuba Large Language Models (LLM ngokufutshane) kuquka neChatGpt edume kakubi. Batyala ukusebenza kwabo kwisixa esikhulu solwazi olusetyenziswa kuqeqesho: okwangoku uqeqesho olulungileyo lufuna ubuncinci iiterabytes ezimbalwa zedatha kwaye inikwe ukuba iterabyte enye ihambelana noonobumba abazibhiliyoni ezingama-90, malunga nezigidi ezingama-75 zamaphepha ombhalo, kulula ukuqonda ukuba kukho. ulwazi oluninzi olufunekayo.
Kodwa ukuba iimodeli azinakwenziwa ubunjineli, kutheni le nto kufuneka sizibuze ingxaki yokwaphulwa kwabucala?
"Nabani na ophambeneyo unokucela ukuba akhululwe kwimishini yokubhabha, kodwa nabani na ocela ukukhululwa kwimisebenzi yenqwelomoya akaphambani." - esekelwe kwinoveli ethi "Catch 22" nguJoseph Heller.
Ukuqokelelwa kwedatha yobukhulu obunjalo bokuvumela ukudalwa kweeprojekthi ezifana ne-ChatGpt okanye ezinye ezifanayo namhlanje lilungelo leenkampani ezinkulu zezizwe ngezizwe ezithe, ngemisebenzi yazo yedijithali, ziye zakwazi ukufumana izandla zabo kwindawo yokugcina ulwazi. kwihlabathi: iWebhu.
UGoogle kunye noMicrosoft, abasele beneminyaka belawula iinjini zokukhangela eziskena iWebhu kwaye zikhuphe ulwazi oluninzi, ngabaviwa bokuqala bokuyilwa kweLLM, ekuphela kwemodeli ye-AI ekwaziyo ukwetyisa ubungakanani bolwazi olufana nezi zichazwe ngasentla.
Kunzima ukukholelwa ukuba uGoogle okanye uMicrosoft angakwazi ukufihla iinkcukacha zobuqu kwidatha yabo ngaphambi kokuba bayisebenzise njengecorpus ekuqeqesheni inethiwekhi ye-neural. Ukungaziwa kolwazi kwimeko yeenkqubo zeelwimi kuguqulela ekuchongweni kwedatha yomntu ngaphakathi kwikhophu kunye nokutshintshwa kwayo ngedatha yobuxoki. Makhe sicinge ngekopus enobungakanani beeterabytes ezimbalwa esifuna ukuqeqesha ngazo imodeli kwaye sizame ukucinga ukuba ungakanani umsebenzi onokuba yimfuneko ukufihla idatha equlethwe ngesandla: kuya kuba yinto engenakwenzeka. Kodwa ukuba sifuna ukuxhomekeka kwi-algorithm ukwenza oko ngokuzenzekelayo, inkqubo ekuphela kwayo ekwaziyo ukwenza lo msebenzi iya kuba yenye imodeli enkulu ngokulinganayo nephucukileyo.
Siphambi kwengxaki yeCatch-22 yakudala: "ukuqeqesha i-LLM ngedatha engaziwa, sifuna iLLM ekwaziyo ukuyichaza, kodwa ukuba sineLLM ekwaziyo ukufihla idatha, uqeqesho lwayo aluzange lwenziwe ngedatha engaziwa. .”
I-GDPR echaza (phantse) emhlabeni jikelele imigaqo yokuhlonipha ubumfihlo babantu, ekukhanyeni kwezi zihloko sele zindala iindaba kunye nokukhuselwa kwedatha yomuntu siqu echaphazelekayo kwisethi yoqeqesho ayicatshangelwa.
Kwi-GDPR, ukusetyenzwa kwedatha yobuqu ngenjongo yokufunda ulungelelwaniso oluqhelekileyo kunye nonxibelelwano kulawulwa ngokuyinxenye kuphela kwiSiqendu 22 esithi: “Isifundo sedatha sinelungelo lokungathotyelwa kwisigqibo esisekwe kuphela ekusetyenzweni okuzenzekelayo, kubandakanywa neprofayili, leyo. ivelisa iziphumo ezisemthethweni kuye okanye ezimchaphazela ngendlela efanayo nebalulekileyo".
Eli nqaku lizisa ukuvinjelwa kwabalawuli bedatha ukuba basebenzise idatha yobuqu yesifundo njengenxalenye yenkqubo yokwenza izigqibo ngokuzenzekelayo eneziphumo zomthetho ngokuthe ngqo kulo mbandela. Kodwa uthungelwano lwe-neural, olulungelelaniswa ngokulula kwiinkqubo zokwenza izigqibo ezizenzekelayo, xa sele luqeqeshiwe lufumana amandla okwenza izigqibo ezizenzekelayo ezinokuchaphazela ubomi babantu. Kodwa ezi zigqibo azisoloko “zisengqiqweni”. Ngexesha loqeqesho, enyanisweni, inethiwekhi ye-neural nganye ifunda ukudibanisa ulwazi kunye nenye, ihlala inxibelelana enye kwenye ngendlela engahambelani ngokupheleleyo. Kwaye ukungabikho kwe "logic" akuwenzi umsebenzi ube lula kumthetho-mthetho ofuna ukuphakamisa ikhaka ekukhuseleni ubumfihlo babantu.
Ukuba umntu uphinde wakhetha ukusebenzisa umgaqo-nkqubo ongqongqo ngokugqithisileyo, umzekelo ukwalela ukusetyenziswa kwayo nayiphi na idatha enovakalelo ngaphandle kokuba ugunyaziswe ngokucacileyo ngumnini, ukusetyenziswa ngokusemthethweni kothungelwano lwe-neural kuya kuba yinto engenakwenzeka. Kwaye ukuyeka ubugcisa benethiwekhi ye-neural kuya kuba yilahleko enkulu, cinga nje ngeemodeli zokuhlalutya eziqeqeshwe ngedatha yeklinikhi yezifundo zabemi abaye bachatshazelwa ngokuyinxenye sisifo esithile. Le mizekelo inceda ekuphuculeni imigaqo-nkqubo yokuthintela ngokuchonga ulungelelwaniso phakathi kwezinto ezikhoyo kwidatha kunye nesifo ngokwaso, ulungelelwaniso olungalindelekanga olunokuthi emehlweni oonompilo lubonakale lungenangqiqo ngokupheleleyo.
Ukubeka ingxaki yokuhlonipha ubumfihlo babantu emva kokugunyazisa ngokungakhethiyo ukuqokelela kwayo iminyaka kubuhanahaniso ukutsho okona kuncinci. I-GDPR ngokwayo kunye nobunzima bayo ijongene neendlela ezininzi zokukhwabanisa ezivumela ukufumana isigunyaziso sokucwangcisa idatha yomntu ngokusebenzisa ukungaqondakali kwamagatya kunye nobunzima bokuqonda.
Ngokuqinisekileyo sifuna ukwenziwa lula umthetho ovumela ukusetyenziswa kwawo kunye nemfundo yokwenyani ekusebenziseni ulwazi lobuqu.
Isiphakamiso sam asikuko ukuvumela iinkampani ukuba zazi idatha yomuntu siqu yabasebenzisi ababhalisela iinkonzo zabo, nokuba bahlawule iinkonzo. Ukusetyenziswa kwedatha yomgunyathi ngabantu babucala kufuneka kwenzeke ngokuzenzekelayo xa besebenzisa iinkqubo ze-intanethi. Ukusetyenziswa kwedatha yangempela kufuneka kugcinwe kwinkqubo yokuthenga yodwa, ukuqinisekisa ukuba ihlala ihluke ngokupheleleyo kwisiseko senkonzo.
Ukwazi incasa kunye nezinto ezikhethwayo zesifundo ngaphandle kokuvumela igama okanye ubuso ukuba inxulunyaniswe nale profayile kuya kusebenza njengendlela yokufihla amagama okwenziwa phezulu okuya kuvumela ngokuzenzekelayo ukuqokelelwa kwedatha kunye nokusetyenziswa kwazo kwiinkqubo ezizenzekelayo ezifana nobukrelekrele bokwenziwa.
Inqaku le Gianfranco Fedele
I-Laravel, edume ngobuchule bayo be-syntax kunye neempawu ezinamandla, ikwabonelela ngesiseko esiluqilima soyilo lwemodyuli. Pha…
ICisco kunye neSplunk zinceda abathengi ukuba bakhawulezise uhambo lwabo lokuya kwiZiko lokuSebenza loKhuseleko (SOC) lekamva nge…
I-Ransomware ilawule iindaba kule minyaka mibini idlulileyo. Uninzi lwabantu luyazi kakuhle ukuba ukuhlaselwa ...
Utyando lwe-ophthalmoplasty kusetyenziswa iApple Vision Pro yombukeli wentengiso lwenziwa eCatania Polyclinic…
Ukuphuhlisa izakhono zemoto ngokufaka imibala kulungiselela abantwana izakhono ezinzima ezifana nokubhala. Ukufaka umbala...
Icandelo lomkhosi wasemanzini ligunya lokwenyani loqoqosho lwehlabathi, elithe lajonga kwimarike ye-150 yeebhiliyoni...
NgoMvulo ophelileyo, i-Financial Times ibhengeze isivumelwano kunye ne-OpenAI. I-FT ikhupha iilayisensi kubuntatheli bayo obukumgangatho wehlabathi…
Izigidi zabantu zihlawula iinkonzo zokusasaza, zihlawula umrhumo wenyanga. Luluvo oluqhelekileyo ukuba…
