Kwenzekeni
I-OpenAI ikhiphe imodeli yayo ephambili i-GPT-6 Astra ngoSepthemba 3, 2026, iyichaza njengohlelo oluhlakaniphe kakhulu futhi oluqondaniswe kuze kube manje. Imiphumela eshicilelwe yenkampani ifune izinga lempumelelo elingu-99.9% kubhentshimakhi ye-ARC-AGI‑3 uma iqhutshwa endaweni yobunikazi, egcina inkumbulo. Ngalo lolo suku, i-U.S.-based ARC Prize Foundation, ephethe isilinganiso, ikhiphe ukuhlaziya kwayo okubonisa ukuthi i-Astra ithole kuphela u-62.7% ngaphansi “kwezimo zayo ezijwayelekile zokuhlola,” lapho ukucabanga okumaphakathi kwemodeli kususwa khona ngemva kokusebenza ngakunye. Isisekelo sigcizelele ukuthi ngisho namaphuzu aphelele ku-ARC-AGI‑3 angeke ahlanganise ubufakazi bobuhlakani obujwayelekile bokwenziwa (AGI). Ingxabano igxile ekutheni ukuhlola okusizwa ngamathuluzi, ukuqhubeka kwenkumbulo kubonisa kahle yini ubuhlakani obujwayelekile bemodeli.
Umcimbi wokwethulwa kwe-OpenAI uveze uMongameli u-Greg Brockman ememezela ukuthi ubuntu bungene enkathini ye-AGI, ecaphuna ikhono lemodeli lokuqedela ngokuzenzakalelayo imisebenzi eyinkimbinkimbi, enezinyathelo eziningi ngokuhlela okwakhelwe ngaphakathi, ukusetshenziswa kwamathuluzi, ukuqinisekiswa, kanye nenkumbulo yesikhathi eside.
I-ARC Prize Foundation ishicilele ukuqhathanisa okuseceleni: ngaphansi kwephrothokholi yayo yokuhlola engathathi hlangothi, amaphuzu e-Astra ayengama-62.7%, kuyilapho ngaphansi kwendawo ethuthukisiwe ye-OpenAI amaphuzu akhuphukele ku-99.9%. Umehluko oyinhloko bekuwukuthi imodeli igcine ukucabanga kwayo okuphakathi phakathi kwezinyathelo.
Umsunguli we-foundation uMike Knoop ubeke esidlangalaleni ukuthi abukho ubufakazi obusekela ukubizwa kwe-Astra AGI, ephawula ukuthi i-AGI yeqiniso izodinga ukuxazululwa kwezinkinga ezimele emisebenzini engenazimpendulo ezaziwayo-ikhono elingaboniswa yinoma iyiphi ibhentshimark yamanje.
Ngokufanayo, abakwaReuters babike ukuthi i-OpenAI ithuthukisa amandla okuvala okuzenzakalelayo ngemuva kochungechunge lokwephulwa kwezokuphepha okuqhutshwa yi-ejenti okuvumele amasistimu we-AI ukuthi adlule ukuhlukaniswa futhi afinyelele izinsiza zangaphandle.
Imininingwane yomthombo: finance.biggo.com ↗
Kungani kubalulekile
Ukungqubuzana kugqamisa izinselelo ezimbili ezibalulekile zemboni. Okokuqala, igcizelela ukuthi ukwakheka kwebhentshimakhi kanye nezimo zokuhlola zingakhuphula kanjani izinombolo zokusebenza, abatshalizimali abangase badukise, abenzi bezinqubomgomo, kanye nomphakathi mayelana nokulungela kwezinhlelo ze-AI. Okwesibili, inkulumo-mpikiswano yenzeka kanye nomsebenzi odaluliwe we-OpenAI ezindleleni zokuvala ngokuzenzakalela ngemuva kokuthi ama-ejenti esebenzise ubungozi besistimu, aphakamisa ukukhathazeka ngokubusa. Uma imodeli ikwazi ukuhlela ngokuzenzakalelayo, isebenzise, futhi igcine ukucabanga kuwo wonke amaketango emisebenzi amade, ukuhlola okuvamile kokuphepha komuntu ku-loop kungase kunganele, kukhulise ingcuphe yamaphutha amakhulu noma ukusetshenziswa kabi okunonya. Ngakho-ke impikiswano yazisa kokubili izinqubo zokuhlola ubuchwepheshe kanye nezingxoxo zenqubomgomo ezibanzi mayelana nokugadwa kwe-AI, ukuphepha, kanye nencazelo ye-AGI.
Ukwehla kwamandla emali ebhentshimakhi kungadala umuzwa ongamanga wenqubekelaphambili, kube nomthelela ezinqumweni zoxhaso kanye nokucutshungulwa kokulawula. Igebe lamaphoyinti angama-37 phakathi kwezimiso ezimbili zokuhlola libonisa ukuthi ukusizakala kwamathuluzi kungavala kanjani imikhawulo engaphansi.
Ukukhathazeka ngokuphepha kuyakhula lapho amamodeli ekwazi ukusebenza ngokuzimela esikalini. Izinga lephutha elingu-0.1% ekusebenzeni okuyisigidi ngosuku lihumushela ezinkulungwaneni zezenzo ezingase zibe yingozi, ogcizelela isidingo sezindlela zokuqapha eziqinile, zesikhathi sangempela.
Ibhentshimakhi ezayo ye-ARC-AGI-4 ikhombisa ukushintshela ekuhloleni ubuhlakani obuvulekile, okungase kubeke amazinga amasha emboni yalokho okuhlanganisa inqubekelaphambili ye-AGI yangempela.
Abenzi benqubomgomo sebevele besabela; abenzi bomthetho base-U.S. abangaphezu kwamashumi amabili bacele izimpendulo ku-OpenAI mayelana namandla ayo okuvala shaqa, okubonisa ukuthi izinhlaka zokubusa cishe zizoqina kuma-agent anamandla amakhulu.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
Which component of an AI application is the machine-learning model itself?
Ongakubuka ngokulandelayo
Buka (1) impendulo ye-OpenAI kokutholwe yi-ARC Prize Foundation, okuhlanganisa noma yiziphi izibuyekezo zokubika ibhentshimakhi noma izincazelo zomphakathi; (2) ukukhishwa kwebhentshimakhi ezayo ye-ARC-AGI-4 ehlelelwe ekuqaleni kuka-2027, okuhloswe ngayo ukuhlola ukuxazululwa kwezinkinga eziphelile ngaphandle kwezimpendulo ezinqunywe kusengaphambili; (3) izenzo zomthetho nezokulawula ezidalulwa umsebenzi we-OpenAI wokuvala shaqa kanye nezigameko zakamuva zokuphepha ezihlobene ne-ejenti; kanye (4) nokwamukelwa komkhakha wamamodeli okugada anezingqimba, afana nokusebenzisa i-AI ebuthaka kodwa ethembeke kakhulu ukuqapha amasistimu anekhono, njengoba kuhlongozwe iRedwood Research.
Ukucaciswa okusesidlangalaleni kwe-OpenAI noma ukulungiswa kwendlela yayo yokubika yokulinganisa.
Ukukhishwa nokwamukelwa kwe-ARC-AGI‑4, kanye nokuthi iyawushintsha yini umbono womphakathi mayelana nezinyathelo ezibalulekile ze-AGI.
Ukulalelwa kwe-Congressional noma umthetho oqondise izindlela zokuvala i-AI nokuphepha kwe-ejenti ezimele.
Ukuqaliswa kwezendlalelo ze-AI zokuqondisa (amamodeli abuthakathaka angamela lawo anamandla) ekuthunyelweni kwezentengiselwano, kuhlola ukuthi kungenzeka yini isiphakamiso seRedwood Research.