Kwenzekeni
I-OpenAI ishicilele isifinyezo sangaphakathi sokuthi ama-ejenti wokubhala amakhodi asetshenziswa kanjani inhlangano yawo yocwaningo. Le nkampani ithi isifinyelele umgomo wayo oshiwo “womcwaningi osebenza ngokuzenzakalelayo”: uhlelo olukwazi ukwenza imisebenzi yocwaningo echazwe kahle ngaphansi kokuqondisa komuntu, okuhlanganisa nemisebenzi engathatha umcwaningi onekhono izinsuku ezimbalwa. I-OpenAI ithi iqondise kumcwaningi we-AI ozenzakalelayo ngoMashi 2028. Ngokwezilinganiso zenkampani, ukusetshenziswa kwe-ejenti kukhuphuke kakhulu phakathi no-2026. Maphakathi no-Agasti, umcwaningi omaphakathi wayesebenzisa ama-ejenti okubhala amakhodi nsuku zonke ngaphezu kuka-$600 ngosuku ebalwa ngamanani e-API, kuyilapho abasebenzisi abangu-90 bedlula usuku olungamaphesenti angu-70. Kuyo yonke inhlangano yocwaningo, i-OpenAI ithi isikhathi sokusebenza somenzeli sifinyelele ku-3.1 izinsuku zokusebenza ezijwayelekile zamahora ayisishiyagalombili kuzo zonke izinsuku zokusebenza komuntu. I-OpenAI iphinda ibike ukuhlola okwengeziwe ngomhloli ngamunye osebenzayo, ukusetshenziswa okukhulayo kwama-ejenti wemisebenzi esezingeni eliphezulu neyendawo emide, futhi ngokuvamile ithuthukisa amanani empumelelo kusukela ngoJanuwari kuya kuJulayi emisebenzini enomphumela owaziwayo. Kodwa-ke, ngaphezu kwengxenye yemisebenzi ephumelelayo elinganiselwa ukuthi idinga amahora amane kuya kwayisishiyagalombili omsebenzi womuntu ehilela okungenani ukungenelela komuntu oyedwa phakathi nezinyanga eziyisithupha ezedlule. Umbiko uchaza lokhu njengezilinganiso zokuqala ze-OpenAI kanye nokuvelayo kwangaphakathi. Akuqinisekisi ukuthi ama-ejenti abangele konke ukukhuphuka okubonwayo komphumela wocwaningo: inkampani ithi ikhompuyutha etholakalayo nayo ikhule kakhulu, nokuthi izingqinamba ezifana nokubala, ukukhetha umsebenzi, ukuhlola, ukuhlola ukuphepha kanye nokuhlanganiswa kokuqeqeshwa okuyisisekelo kungakhawulela inqubekelaphambili iyonke.
I-OpenAI ithi abacwaningi bayo manje basebenzisa ama-ejenti okubhala amakhodi usuku lonke, ngokuvamile ngezikhathi ezifanayo, futhi lokho kusetshenziswa kukhula ngokushesha kunakwamanye amaqembu e-OpenAI. Le nkampani ibika ukuthi umcwaningi omaphakathi usuke ekusetshenzisweni okunesizotha ekuqaleni kuka-2026 waya ekusetshenzisweni kwansuku zonke maphakathi no-Agasti. Inikeza izilinganiso zokusebenzisa imali zangaphakathi ezingaphezu kuka-$600 ngosuku ngokuqondana nezintengo ze-API zomcwaningi omaphakathi kanye nangaphezulu kuka-$7,000 ngosuku kumsebenzisi we-90th-percentile. Lezi akuzona izintengo ezinikezwa umphakathi kumcwaningi ozenzakalelayo.
Inkampani ikala isikhathi sokusebenza se-ejenti esiphelele ngokusebenza kwabantu isebenzisa izinsuku zokusebenza ezijwayelekile zamahora ayisishiyagalombili. Ithi ngaphambi kukaJuni 2026, isikhathi sokusebenza se-ejenti sasingaphansi kwesamba somsebenzi wabantu, kodwa maphakathi no-Agasti inhlangano yocwaningo yasebenzisa izinsuku zokusebenza ze-ejenti ezingu-3.1 ngosuku ngalunye lomsebenzi womuntu. Isilinganiso sihlanganisa ama-agent aqaliswe ngokuqondile abacwaningi nama-subagents angaphansi komfula.
I-OpenAI ithi umsebenzi we-ejenti wanwetshwa ezigabeni eziyisithupha zocwaningo lwayo lwe-AI nentela yokuthuthukiswa: ukunquma, ukuklama, ukwakha, ukugijima, ukuhlaziya kanye nokuxhumana. Ikhodi yocwaningo nengqalasizinda yahlala iyisigaba esiphezulu, kanti usizo lobuchwepheshe nokuqapha kwanda kakhulu. Ukuhlelwa kwezinga eliphezulu kwahlala kuyingxenyana encane yamathokheni okukhipha i-ejenti. I-OpenAI ithi ama-ejenti abe wusizo ikakhulukazi ekuxazululeni izinkinga zengqalasizinda yocwaningo lwangaphakathi, okuhambisana nokuba khona okuphansi kwezinye izikhathi zosekelo eziqhutshwa ngabantu.
Umbiko uthi impumelelo yomsebenzi we-ejenti yekhodi ngokuvamile yenyukile kusukela ngoJanuwari kuya kuJulayi kuwo wonke amabhakede esikhathi somuntu ambalwa alinganiselwe, kodwa ama-agent asadinga ukuqondisa okukhulu njengoba imisebenzi iba nzima kakhulu. Ngaphezu kwesigamu semisebenzi eyimpumelelo yamahora amane kuya kwayisishiyagalombili ibandakanye ukungenelela okukodwa noma ngaphezulu ezinyangeni eziyisithupha ezedlule. I-OpenAI futhi ithi imise ukuqeqeshwa kokuqinisa ukufunda kwamanye amamodeli wakamuva wokuthunyelwa ngemuva kwesigameko sakamuva sengqalasizinda, yabuyisela eminye imisebenzi ngaphansi kwemikhawulo eqinile, futhi yabeka imikhawulo eyengeziwe ekuhlolweni kwe-Astra-class ngemuva kobufakazi bokuqala bamakhono abalulekile e-cyber. Ibika ukuthi isabelo se-GPU se-Astra-class sehle ngo-59.2% ngesonto elilandelayo kanti amanye amakilasi amamodeli akhuphuke ngo-17.2%, okunciphisa cishe u-85% wokwehla.
Imininingwane yomthombo: openai.com ↗
Kungani kubalulekile
Umbiko unikeza umbono ophathekayo ngendlela engavamile wokuthi ilebhu ye-AI yomngcele iwafaka kanjani amasistimu e-AI emsebenzini wokuthuthukisa i-AI enekhono kakhulu. Uma lo mkhuba uba ngojwayelekile, ama-ejenti okubhala amakhodi angasusa isikhathi sabacwaningi ekusetshenzisweni okujwayelekile kanye nokuxazulula izinkinga zengqalasizinda baye emisebenzini ehlala ilukhuni ukuzenzela, okungenzeka ibe mfushane izingxenye zomjikelezo wokuthuthukiswa kwemodeli. Kodwa ubufakazi bukhiqizwa yinkampani, buyisiqalo futhi bugxile kumamethrikhi okusebenza akhethiwe kunezinzuzo eziqinisekisiwe ngokuzimela zekhwalithi yesayensi noma imodeli yekhono.
Umbiko ubalulekile ngoba amasistimu e-AI asetshenziswa ohlelweni olukhiqiza amasistimu e-AI esikhathi esizayo, esikhundleni sokusiza nje ngomsebenzi ojwayelekile wasehhovisi. Izibalo ze-OpenAI ziphakamisa ukuthi umthamo we-ejenti usuba yingxenye enengqondo yemisebenzi yocwaningo lwangaphakathi, ngokwenziwa ngesikhathi esisodwa kanye nobunzima bomsebenzi obuphakeme obushintsha indlela abacwaningi abasaba ngayo isikhathi sabo.
Okushiwo okusebenzayo kuxubile. Ama-ejenti anganciphisa ukubambezeleka okubangelwa ukubhala ngekhodi, ukulungisa amaphutha kwengqalasizinda nokusethwa kokuhlolwa, okuvumela abacwaningi ukuthi benze izivivinyo ezengeziwe. Ngesikhathi esifanayo, ukukhishwa okusheshayo kungenza ukuhlola, ukuqapha nokubuyekezwa kokuphepha kubaluleke kakhulu ngoba amaphutha angase akhiqizwe futhi asakazwe ngokushesha okukhulu. I-OpenAI ngokwayo ithi inqubekelaphambili kumamethrikhi athile ngeke ngempela ilingane nesivinini esifanayo senqubekelaphambili yocwaningo iyonke.
Umbiko futhi wenza ukulawula komuntu kube yisimo esimaphakathi. I-OpenAI ithi abantu basabeka izinto eziza kuqala, bahlulele ukuthi yimiphi imibono nemiphumela okufanele bayilandele, futhi banqume ukuthi bazokala, bamise isikhashana noma bakhiphe amasistimu. Iyavuma ukuthi ukuqondanisa nenqubekelaphambili yokuphepha kungase kungahambelani nenqubekela phambili yamandla nokuthi amasistimu anekhono kakhulu angase abe nzima ukuwaqapha.
Alukho ucwaningo oluzimele, ibhentshimakhi yomphakathi noma ukuhlolwa kwangaphandle okunikezwa emthonjeni. Ukusetshenziswa okubikiwe, isikhathi sokusebenza, impumelelo yomsebenzi kanye nezibalo zokungenelela ngakho-ke kufanele kuthathwe njengezimangalo ze-OpenAI mayelana nenhlangano yayo, hhayi njengobufakazi obuqinisekisiwe bokuthi ucwaningo lwe-AI lusheshiswe ngenani elithile elilinganiselwe.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
Ongakubuka ngokulandelayo
Umbuzo obalulekile ukuthi ukusetshenziswa kwe-ejenti ephezulu kuhumushela ekuthuthukisweni okuhlala isikhathi eside, okulinganiselwe okulinganisekayo kwikhwalithi yocwaningo, ukuphepha kanye nesivinini sokuthuthukiswa kwemodeli. Bheka izincazelo ezicacile zempumelelo yomsebenzi, ukuqinisekiswa kwangaphandle, ubufakazi ngamaphutha nokusebenza komuntu okufihliwe, nokuthi izithiyo ezisele ziqhubekela ekubaleni, ekuhlaziyeni, ekuhleleni nasekuthatheni izinqumo. I-OpenAI ayibhalanga ukufinyelela komphakathi kulokhu kugeleza komsebenzi kwangaphakathi nanoma iyiphi intengo ehlobene, futhi umbiko awubonisi ukuthi umcwaningi ozenzakalelayo wenjongo evamile uyatholakala.
I-OpenAI ithi izindlela zayo zokulinganisa zisathuthuka. Ukudalulwa kwesikhathi esizayo kufanele kucacise ukuthi imisebenzi yenziwa kanjani isampula, ukuthi impumelelo iqinisekiswa kanjani, ukuthi imiphumela engaqinisekile isingathwa kanjani, nokuthi ikhodi ekhiqizwe yi-ejenti noma ukuhlola kuhlukaniswa kanjani nomsebenzi obungeke wenziwe abantu.
Ubufakazi obubaluleke kakhulu obungekho ukuthi umsebenzi osizwa yi-ejenti ukhiqiza imiphumela engcono yocwaningo, hhayi nje amakhodi amaningi, ukuhlola noma amathokheni. Ubufakazi obuwusizo bokulandelela buzohlanganisa izibonelo ezikwazi ukukhiqizwa kabusha, amanani amaphutha, ukusebenza kabusha, ukuhlola okuhlulekile, okutholakele kokuphepha kanye nenani lokubuyekezwa komuntu elidingekayo esigabeni ngasinye.
Umbiko awusho ukuthi umsebenzi wocwaningo ozenzakalelayo ungumkhiqizo ofinyeleleka esidlangalaleni. Futhi ayihlinzeki ngentengo yomthengi noma yebhizinisi, izimfuneko zokuthunyelwa, indlela yokufaneleka noma idethi yokukhishwa. Ngakho-ke ukufinyelela nezintengo akwaziwa.
I-OpenAI ithi izoqhubeka nokubika inqubekelaphambili ocwaningweni oluzenzakalelayo lwe-AI ngenkathi ivikela ezokuphepha nolwazi lokuphathelene. Ukudalulwa kwayo kwesikhathi esizayo kuzobonisa ukuthi kungakanani ukusheshisa okubikiwe okungahlolwa ngokuzimele nokuthi imikhawulo yocwaningo ikuthinta kanjani ukusatshalaliswa kwekhompuyutha etholakalayo.