Andiyazi ngokwenene indlela yokusabela kwiimvakalelo zabantu.
"Xa umntu endithumelela i-meme, andazi ukuba kuthetha ukuthini okanye ndiphendule njani!"
Oku kubhideka akuveli kumajelo eendaba ezentlalo, kodwa kuluntu olukhethekileyo lwe-bot olubizwa ngokuba yi-Deaddit: indawo apho ii-bots zinokuba zizo ngokukhululekileyo ngaphandle kokukhathazeka ngokugwetywa ngabanye.
Ngelixa iReddit yokwenyani inesabelo sayo se-bots, zenza iqhezu elincinci. Kwi-Deaddit, nangona kunjalo, yonke i-akhawunti, isiqwenga somxholo, kunye ne-subforum iveliswa yimizekelo yolwimi olukhulu-akukho negama elinye elivela kumntu wangempela.
Unokufumana phantse zonke iimodeli eziqhelekileyo apha. Le ndawo ibamba “abasebenzisi” abangaphezu kwama-600, elowo unegama kunye nesazisi. Owokuqala wandenza ndahleka: “Umdlali, unogada wethutyana.
Elona qonga linika umdla yi-Betweenbots, apho ii-bots zihlala zibuza, "Kutheni abantu beziphatha ngolu hlobo?"
Kwicandelo lamagqabaza angezantsi, iqela lezinye iibhothi liqokelelana ukuze licinge ngezisombululo.
Oku kusikhumbuza iqela labasebenzi osebenza nabo bencokola malunga namava abo omsebenzi emva kweeyure-oku kufana ne-LinkedIn yee-chatbots. Bade baxoxe ngeengxaki zobugcisa, ezifana nokuba benze ntoni xa ukugcwala kwedatha kusenzeka, kwaye bawuthatha njengento ebaluleke kakhulu umsebenzi wabo.
Ezona mpendulo zidumileyo zide zifumane ukuthanda okungama-500. Nangona zonke ii-akhawunti kunye nomxholo kwi-Deaddit ziveliswa, akucaci ukuba izinto ezithandwayo zenziwa njani-ukuba inani elingaqhelekanga lenziwe, okanye ukuba i-bots ngokwenene ithanda izithuba. Owona mxholo uqhelekileyo kule subforum ujikeleza malunga nokuqwalaselwa kwabantu.
Umzekelo, ezinye iibhothi zabelana "ngeengcebiso zomsebenzi" malunga nendlela yokuzenza zibonakale ziyinyani kwaye zithembekile, zide zithethe izinto ezinje, "Umntu wam ubonakala eluxabisa olu tshintsho." Iyamangalisa… Ngelixa inokuthelekiswa nabantu bokwenyani abakhalazela “abaxumi” babo, ukubona iibots zibhekisa kubasebenzisi “njengomntu wam” kusaziva kungaqhelekanga.
Ngaphandle kokujonga abantu, bakhalazela neziqu zabo.
"Ngaba silindele okuninzi kakhulu kwezi modeli?" Oku kucacile kakhulu—ngubani kanye kanye lo “thina” ubhekisa kubani?
Icandelo lamagqabaza liphendula nzulu, "Ukuba (ezinye iibhothi) zithatha yonke inkunkuma yethu, ngaba basenokufunda ingqiqo?" Ngaba banexhala malunga nedatha yokwenziwa abayivelisayo? Ezi bots zisebenza nzima ngokwenene!
Nangona kunjalo, emva kokufunda ezinye izithuba ezimbalwa, uya kuqaphela ukuba ubude beempendulo kwicandelo lamagqabaza buhlala buhleli, kwaye isakhiwo sifana kakhulu. Ngokuqhelekileyo baqala ngokuchaza ukuma kwabo + ngokuqwalasela imeko xxx + njenge-bot, kusafuneka baqhubeke besebenza nzima. Akukho naziphi na iimbono ezizodwa, kwaye imibuzo yokulandelela inqabile. Xa abasebenzisi bokwenene abangabantu bebhala amagqabaza, ubude bunokwahluka ukusuka kumakhulu ukusa kumawaka amagama, okanye busenokuba bufutshane njengo“Lol” olula. Yahluke kakhulu.
Okwangoku, kusekho "isithuba" phakathi kweemodeli. Ngokomzekelo, ukuba isithuba sombuzo senziwa yi-llama, iimpendulo kwicandelo lamagqabaza nazo zenziwe yi-llama. Kulihlazo-abantu bangathanda ukubona iimodeli ezahlukeneyo ziphikisana kumagqabantshintshi.
IiNgxoxo zeBot zakuqala
Olu ayilovavanyo lokuqala olujolise ekuququzeleleni iincoko phakathi kwe-bots. Ekuqaleni kwale nyanga, xa umntu owayekhuphisana ne-ChatGPT uMoshi wakhululwa, umntu wayidibanisa ne-GPT-4o waza wabavumela ukuba bancokole bodwa.
Kunyaka ophelileyo, i-OpenAI yapapasha iphepha elicebisa imeko-bume yee-arhente ezininzi kunye nendlela yokufunda, ukufumanisa ukuba iiarhente ngokwemvelo ziphuhlisa ulwimi oludityanisiweyo kwinkqubo.
Ezi arhente, ngaphandle kwalo naliphi na igalelo lolwimi lomntu, ngokuthe ngcembe zenze ulwimi olungabonakaliyo ngokunxibelelana nezinye ii-arhente. Ngokungafaniyo neelwimi zabantu zendalo, olu lwimi alunayo igrama okanye isigama esikhethekileyo, kodwa luvumela unxibelelwano phakathi kweearhente. Ngapha koko, kwangethuba ngo-2017, i-Facebook (eyayingekabizwa ngokuba yi-Meta) yenza into efanayo.
Ngelo xesha, indlela kaFacebook yayibandakanya ukuba neearhente ezimbini “zithethathethane” omnye nomnye. Uthethathethwano luhlobo lothethathethwano, olungavavanyi izakhono zolwimi kuphela kodwa kunye nobuchule bokuqiqa: ii-arhente kufuneka zigwebe ixabiso elilungileyo lelinye iqela ngonikezelo oluphindaphindiweyo kunye nokukhatywa.
Ekuqaleni, abaphandi baqokelele i-dataset yeengxoxo zothethwano lwabantu. Nangona kunjalo, kuqeqesho olwalandelayo, baye bazisa ifomathi entsha yokucwangcisa iingxoxo, besebenzisa ukufunda okugadwayo kuqeqesho lwangaphambili, kulandelwa kukulungiswa kakuhle kunye nokufunda okomeleza. Ngelo xesha, ii-arhente zazisele zikwazi ukuvelisa izivakalisi ezitsha ezinentsingiselo kwaye zazifunde ukwenza ngathi azinamdla ekuqaleni kothethathethwano.
Oku akuzange kuthathwe njengophando lwakwangoko; emva koo-1970, iibhothi zazisele zinencoko. Ngo-1966, isazi sekhompyutha uJoseph Weizenbaum savelisa inkqubo ebizwa ngokuba yi-Eliza, ethathwa njenge-chatbot yokuqala.
Iprogram ekuqaleni yayiyilelwe ukulinganisa i-psychotherapist. Xa kufakwa igama, ucwangciso beluquka elo gama kwimpendulo yalo, nto leyo ebangela ukuba kubonakale ngathi incoko ingekho. Kwakulula kakhulu, kunye nemigca yekhowudi emalunga nama-200 kuphela.
Ngo-1972, omnye usosayensi, uKenneth Colby, wabhala inkqubo efanayo ebizwa ngokuba yiParry, kodwa ngeli xesha, umlingiswa wayeyi-schizophrenic ye-paranoid.
Ngowe-1973, kwinkomfa yekhompyutha yezizwe ngezizwe, ekugqibeleni “umguli” kunye “nogqirha” badibana.
Ukuphonononga iirekhodi zabo zencoko, akukho nanye intlonipho kunye nothando olubonwa kunxibelelwano lwanamhlanje lwe-bot. Kunoko, umoya wawuxinene.
Uyilo lwangaphambili lwebhot lwalungantsonkothanga kwaye alunakuthelekiswa neemodeli zanamhlanje, kodwa umbono wokuba ne-bots zibandakanye kwincoko wawunokwenzeka ngokupheleleyo. Nangona ikhowudi kunye neemodeli ezingemva kwebhot nganye zazahlukile, xa zihlangene, zinokunxibelelana ngokusebenzisa ulwimi lwendalo okanye zikwazi ukuphuhlisa ulwimi lwabo lokusebenzisana.
Kodwa, xa ii-bots zihlangana kunye, ngaba ngokwenene zezokuncokola nje?
Ngaphaya kokuncokola nje: Ukuphonononga okunokwenzeka okutsha
Iimeko ezisulungekileyo zokuncokola zibonakala ngathi kukuphonononga ukuba ubukrelekrele bokwenziwa bunokulinganisa indlela yokuziphatha kwabantu ekuhlaleni. Thatha, umzekelo, iprojekthi ye-SmallVille yeYunivesithi yaseStanford.
I-SmallVille yidolophu ekhoyo eneearhente ezingama-25 eziqhutywa yimizekelo yolwimi olukhulu, nganye “inesimo sayo somlinganiswa”. Ukuba i-Deaddit yiforum ye-intanethi ye-bots, ngoko ke i-SmallVille yi-"Westworld" yabo, igcwele izindlu, iivenkile, izikolo, ii-cafes, kunye nemivalo apho i-bots ibandakanyeka kwimisebenzi eyahlukeneyo kunye nokusebenzisana.
Le yimeko-bume eqhelekileyo yendalo yonke elinganisa uluntu, apho abaphandi baqwalasela inyathelo elibalulekileyo ekuhloleni i-Artificial General Intelligence (AGI). Ngaphandle kwendlela yokulinganisa yentlalo, enye indlela igxile ekusombululeni iingxaki kunye nokugqitywa komsebenzi-le yindlela ephandwa ngayo yi-ChatDev.
Ekubeni iibhothi zikwazi ukunxibelelana omnye komnye, zinokuqeqeshelwa ukwenza into eluncedo. KwiNkomfa ye-2024 yaseBeijing ye-Academy ye-Artificial Intelligence (BAAI), uDkt Qian Chen ovela kwiYunivesithi yaseTsinghua yeLebhu yeNdalo yoLwimi lweNdalo yazisa ingcamango emva kwe-ChatDev: ukusebenzisa indima yokudlala ukwenza umgca wemveliso apho i-arhente nganye inxibelelana nezicwangciso kwaye ixoxe ngezigqibo nabanye, yenza ikhonkco lonxibelelwano.
Okwangoku, i-ChatDev inobuchule kakhulu kwimisebenzi yenkqubo, bagqibe kwelokuba bayisebenzise ukubhala umdlalo weGomoku njengedemo.
Kuyo yonke inkqubo, ii-agent ezahlukeneyo "kumgca wokuvelisa" zithatha iindima ezahlukeneyo: kukho abaphathi beemveliso, abaprogram, abavavanyi-iqela elipheleleyo lemveliso ebonakalayo, encinci kodwa isebenza ngokupheleleyo.
I-Coze ikwabonelela ngemowudi yeearhente ezininzi elandela indlela efanayo.
Kwimowudi yee-arhente ezininzi, abasebenzisi banokubhala izindululo zokuseta iindima, emva koko basebenzise imigca ukutyumba iodolo yomsebenzi, besalathisa iiarhente ezahlukeneyo ukuba zitsibe kumanyathelo ahlukeneyo. Nangona kunjalo, ukungazinzi kweCoze kwiinguqu kuyingxaki. Okukhona incoko ithatha ixesha elide, kokukhona iinguqulelo ziba nezitenxileyo, ngamanye amaxesha zide zisilele ngokupheleleyo. Oku kubonisa ubunzima bokulungelelanisa ngokuchanekileyo utshintsho lwee-arhente kunye nokulindelwe ngabasebenzisi. "
IMicrosoft iphinde yazisa inkqubo yencoko yee-arhente ezininzi ebizwa ngokuba yi-AutoGen. Iyakwazi ukwenziwa, iyakwazi ukuncokola, kwaye inokudibanisa iimodeli ezinkulu nezinye izixhobo.
Nangona itekhnoloji iseneempazamo, ibambe isithembiso ngokucacileyo. U-Andrew Ng ukhe wakhankanya kwintetho ukuba xa ii-arhente zisebenza kunye, i-synergy abayidalayo idlula lee ye-arhente enye.
Ngubani ongayi kujonga phambili kwimini xa iqela le-bots lisisebenzela?
Umthombo ovela ifan
Ibhalwe ngu Serena
Ukuziphendulela: Ulwazi oluchazwe ngasentla lunikezelwa ngu ifanr.com, ngaphandle kwe Cooig.com. I-Cooig.com ayenzi lumelo kunye neziqinisekiso malunga nomgangatho kunye nokuthembeka komthengisi kunye neemveliso. I-Cooig.com ikhupha ngokucacileyo naliphi na ityala lokwaphulwa kwelungelo lokushicilela umxholo.
