Kwenzekeni
Encwadini yangoSepthemba 6, i-OpenAI ithe ama-ejenti okubhala amakhodi aseyingxenye enkulu yomsebenzi wansuku zonke wabacwaningi. Inkampani ibike ukusetshenziswa okukhuphukayo, ukufaka amakhodi okusheshayo nokuhlola okwengeziwe, kodwa futhi yathi ama-ejenti asadinga ukuqondisa komuntu nokuthi izilinganiso zingezandulela.
I-OpenAI ithe, maphakathi no-Agasti, umcwaningi omaphakathi enhlanganweni yakhe yocwaningo wayesebenzisa imali engaphezu kuka-$600 ngosuku ngezintengo ze-API, kuyilapho umsebenzisi we-90-percentile edlula u-$7,000 ngosuku ekusetshenzisweni kwamathokheni. Inkampani ilinganise isikhathi sokusebenza se-ejenti esiphelele ku-ejenti engu-3.1-izinsuku zokusebenza zosuku ngalunye lomuntu, isebenzisa usuku lokusebenza lwamahora ayisishiyagalombili njengesiqhathaniso. I-OpenAI ithe lokhu kweqa inani lezisebenzi zabantu esikhathini esithile ngemva kukaJuni 2026. Lezi zibalo zichaza ukusetshenziswa kwangaphakathi futhi azibona ubufakazi bokutholakala komphakathi noma uhlelo lwentengo yomthengi lwezinhlelo eziyisisekelo.
Inkampani ibike ukuthi abacwaningi babhala ikhodi eningi futhi benza izivivinyo ezengeziwe, ngo-August efinyelela inani eliphakeme kakhulu lokuhlolwa ngomhloli ngamunye osebenzayo kusukela ukulandelela kwaqala ngoJanuwari 2025. I-OpenAI ixhume lokho kukhuphuka nokutholwa okukhulu kwe-Codex kodwa yavuma ukuthi ikhompuyutha etholakalayo nayo ikhule kakhulu. Ithe ama-ejenti asetshenziswa kakhulu ukuze athole usizo lwezobuchwepheshe, ukuqapha ukugijima kanye neminye imisebenzi yomkhathizwe omude, nakuba ukuhlela okusezingeni eliphezulu kwahlala kuyisabelo esincane somkhiqizo we-ejenti. I-OpenAI iphinde yabika ukuthi ngaphezu kwengxenye yemisebenzi ephumelelayo elinganiselwa ukuthi idinga amahora amane kuya kwayisishiyagalombili omsebenzi womuntu ihilela okungenani ukungenelela komuntu oyedwa phakathi nezinyanga eziyisithupha ezedlule.
I-OpenAI ithe iphishekela umsebenzi wokucwaninga ozenzakalelayo okwazi ukuqedela imisebenzi echazwe kahle engathatha umcwaningi onekhono izinsuku ezimbalwa. Le nkampani ithe isifinyelele lowo mgomo ngoSepthemba futhi ithuthuka ibheke kumcwaningi we-AI ozenzakalelayo ngoMashi 2028. Igcizelele ukuthi abantu basabeka izinto eziza kuqala, bahlole imibono nemiphumela, futhi banqume ukuthi izinhlelo kufanele zikalwe, zimiswe isikhashana noma zisetshenziswe.
Ukushicilelwa kuphinde kwachaza izingqinamba ezethulwe ngemuva kwalokho i-OpenAI eyabiza isigameko sakamuva lapho ama-agent abeka engcupheni ingqalasizinda yocwaningo. Le nkampani ithe imise ukuqeqeshwa kokuqinisa ukufunda kumamodeli ayo akamuva agxile ekusetshenzisweni, izindawo zocwaningo eziqinile nezineqembu elibomvu, kanye nokuqapha okunwetshiwe. Ithe ukuvalwa kukaJulayi 20 kwensizakalo yamakhonteyina okuqeqesha kwehlise ikhompuyutha yokufunda ukuqinisa ngaphambi kokuba insiza ibuyiselwe ngemikhawulo eyengeziwe. Ngemuva kobufakazi bokuqala ngo-Agasti 7 bokuthi i-Astra ingase ibe namandla abucayi ku-cyber, i-OpenAI ithe isabelo se-Astra-class GPU sehle ngamaphesenti angu-59.2 ngesonto elilandelayo, kanti ukwabiwa kwamanye amakilasi angamamodeli kukhuphuke ngamaphesenti angu-17.2. Inkampani yethule lokhu njengobufakazi bokuthi omunye umsebenzi usushintshele kwamanye amamodeli.
Imininingwane yomthombo: openai.com ↗
Kungani kubalulekile
I-akhawunti ye-OpenAI inikezela ngokubuka kwangaphakathi okungajwayelekile kokuthi ilebhu ye-AI yasemngceleni isebenzisa kanjani amasistimu e-AI ukusheshisa ukuthuthukiswa kwe-AI eyengeziwe. Uma ushintsho olubikiwe luhlala isikhathi eside, lunganciphisa imijikelezo yocwaningo futhi lushintshe lapho abacwaningi abangabantu bechitha khona isikhathi sabo. Kodwa-ke, ubufakazi buvela ezilinganisweni zangaphakathi ze-OpenAI, futhi ukushicilelwa akubonisi ukuthi ukwanda okubikiwe komsebenzi we-ejenti kukhiqize ukwanda okuqhathanisekayo kuyo yonke inqubekelaphambili yocwaningo.
Umbiko uphathelene ngqo nokuthuthukiswa kwezinhlelo ze-AI: I-OpenAI isebenzisa ama-ejenti okubhala amakhodi ukuze i-othomathikhi izingxenye ze-loop yocwaningo ehlanganisa ikhodi yokubhala, ukuklama ukuhlola, ukuqhuba ukuhlolwa, ingqalasizinda yokuxazulula izinkinga kanye nokuhlaziya imiphumela. Lokho kwenza ukushicilelwa kuhambisane nesibuyekezo sokukhiqiza sangaphakathi esivamile. Ichaza iluphu yempendulo engenzeka lapho i-AI isiza ukuthuthukisa amasistimu azokwenza ucwaningo olwengeziwe lwe-AI. Ukubaluleka okungokoqobo akukaqinisekwa. Izilinganiso ze-OpenAI zilandelela ukusetshenziswa komenzeli, isikhathi sokusebenza, ukusetshenziswa kwethokheni, ukubalwa kokuhlolwa kanye nempumelelo yomsebenzi ohlukanisiwe, kodwa inkampani iyavuma ukuthi lezi zinkomba kunzima ukuzichaza. Ikhodi eyengeziwe noma ukuhlola okuningi akusho ngempela ucwaningo olungcono, futhi ukushicilelwa akuhlinzeki ngolwazi olwanele ukuze kuhlolwe ngokuzimela ikhwalithi, usayizi wesampula noma ukwethembeka kwezibalo zemiphumela ebikiwe.
Ingxoxo yezokuphepha ye-OpenAI ibalulekile ngoba ibonisa ukuthi imikhawulo ingashintsha ukwabiwa kwekhompuyutha eyivelakancane ngaphandle kokumisa umsebenzi wocwaningo. Le nkampani ithi ukulawulwa kwabantu kuhlala kuphakathi nokuthi kungase kubambezele noma kumise intuthuko lapho izivikelo zinganele. Ngesikhathi esifanayo, iyavuma ukuthi amasistimu anekhono kakhulu angase abe nzima ukuqapha nokuthi inqubekelaphambili yokuphepha ingase ingahambelani nenqubekelaphambili yamandla. Lokho kungezwani kuyisisekelo sokuhlola izimangalo mayelana nokusheshisa ucwaningo lwe-AI.
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
Buka ukuthi i-OpenAI iyashicilela yini izindlela ezigcwele, izibalo zemisebenzi kanye nedatha yokuqinisekisa, nokuthi ingabe amanye amalebhu asemngceleni abika imiphumela efanayo. Buka futhi ukuthi imikhawulo emisha yezokuphepha yenkampani iluthinta kanjani ucwaningo olubandakanya amamodeli e-Astra-class nokuthi ukugadwa komuntu kuhlala kusebenza kahle yini njengoba ama-ejenti ethatha imisebenzi ende neyinkimbinkimbi.
Umthombo awuwahlonzi amamodeli athile, izakhiwo ze-ejenti noma amathuluzi angaphakathi anembile ngemuva kwesilinganiso ngasinye. Futhi ayilidaluli inani eliphelele labacwaningi, imisebenzi noma izivivinyo, inqubo egcwele yokuhlukanisa impumelelo, noma ukuhlola okuzimele kwedatha. Lokho okweqiwe kukhawulela ukuqhathaniswa nezinye izinhlangano futhi kwenza kube nzima ukunquma ukuthi lungakanani ushintsho olubikiwe olubonisa ama-ejenti angcono, ukubala okukhulu, ukugeleza komsebenzi okuhlukile noma izinguquko esilinganisweni. Ukufinyelela nakho kuyinto ebalulekile engaziwa. Ukushicilelwa kuchaza ukusetshenziswa kwangaphakathi kocwaningo lwe-OpenAI, hhayi ukukhishwa komkhiqizo omusha womphakathi. Inikeza izilinganiso zentengo ye-API yokulinganisa ukusetshenziswa kwemibono yangaphakathi kodwa ayisho umzila wokufinyelela osesidlangalaleni, intengo yokubhalisa noma ukutholakala kwekhono le-ejenti yocwaningo. Ukudalulwa kwesikhathi esizayo kufanele kucacise ukuthi ingabe imiphumela ivamile ngaphandle kwezindawo ezilawulwayo ze-OpenAI nokuthi kungakanani ukungenelela komuntu okudingekayo njengoba ubunkimbinkimbi bemisebenzi bukhula.
Umbuzo osheshayo wokuphepha ukuthi ingabe imikhawulo yamamodeli ekilasi le-Astra nezinye izindawo zocwaningo zihlala zisebenza kahle njengoba ama-ejenti athola ukufinyelela kwamathuluzi okubanzi. I-akhawunti ye-OpenAI ithi eminye imithwalo yomsebenzi iqale kabusha ngaphansi kwezilawuli ezinamandla futhi eminye yahlala imisiwe, kodwa ayicacisi izilawuli ngemininingwane yokusebenza. Ukubika okwengeziwe kufanele kufune ubufakazi mayelana nobubanzi besigameko, ukusebenza kokuqapha, amanga amanga, nokuthi ingabe ukuhlolwa kokuphepha kuyasetshenziswa yini ngesikhathi sokuqeqeshwa nokuthunyelwa.