Kwenzekeni
Ibhentshimakhi ye-Epoch AI's FrontierMath manje ifaka kuhlu inkinga yesitayela se-Apéry‑irrationality‑njengobufakazi njengoba isixazululiwe, ibeka ubudlelwano bomuntu-AI obufakazele ukungabi nangqondo kwe-ζ(5) kune-ζ(3) yakudala. Isixazululo sishicilelwe njengokuphrinta kwangaphambili okune-Lean formations, kodwa inkampani iphawula ukuthi ubufakazi abuyilandeli ifomethi yesitayela se-Apéry ibhentshimakhi ebidingeka ekuqaleni. Ukuze ithwebule le nuance, i-Epoch AI yethule ilebula ethi “human + AI” ngoSepthemba 16 ezimweni lapho umuntu eqondisa imodeli ngezinyathelo eziphindaphindayo ezingeke ziphumelele ngokuzenzakalelayo.
Inkundla ye-FrontierMath ye-Epoch AI, eyakhelwe ukumelana nokukhumbula ngekhanda nokuqhathanisa nephethini, manje isiqopha izinkinga ezivuliwe ezingu-49, ezinezimpawu eziphawuliwe eziyisishiyagalolunye—ezine zizimele ngokugcwele neziyisihlanu ngaphansi kwesigaba esisanda kwakhiwa esithi “human + AI”. Inkinga yesitayela se-Apéry ijoyina iqembu lakamuva ngemuva kokuthi umzamo wokuhlanganyela ukhiqize ubufakazi bokungabi nangqondo kwe-ζ(5). Isixazululo sabhalwa ngokuphrinta kwangaphambili okuhlanganisa ukwenziwa kube semthethweni kwe-Lean, kodwa ababhali bayavuma ukuthi ubufakazi abufani nesakhiwo sokuphindaphinda somugqa wesitayela se-Apéry ibhentshimakhi eyayifunwe ekuqaleni.
Ushintsho lwenqubomgomo yenkampani ngoSepthemba 16 lwethule ilebula ethi “human + AI” ukuze ithwebule izimo lapho umuntu eqondisa imodeli ngokuphindaphindiwe, inqubo okungenzeka ukuthi ibingeke iphumelele ngaphandle kwaleso siqondiso. Ikhasi le-Epoch AI uqobo libala inkinga njengexazululiwe naphezu kokungafani, igcizelela ukubaluleka kwelebula ukugwema ukuchazwa kabi kwamakhono azimele e-AI.
Lesi simemezelo silandela ezinye izenzakalo zakamuva ze-FrontierMath, ezifana nokugcwala kwe-GPT-6 Astra eduze kwe-Tier 4 (kubikwe ngokunemba okungu-97.6 %) kanye nesixazululo “somuntu + AI” senkinga yokhetho olusekelwe ekomitini. Leyo ntuthuko enkulu igcizelela indima yebhentshimakhi ekulandeleleni kokubili ukucabanga okuzimele nokusizwa kwe-AI.
Imininingwane yomthombo: startupfortune.com ↗
Kungani kubalulekile
Isimemezelo sibonisa ukuthi abaklami bebhentshimakhi bazivumelanisa kanjani nezimo ukuze bahlukanise ukucabanga kwangempela kokuzimela ekusetshenzisweni kwemodeli okusizwayo, umehluko obalulekile ekuhloleni inqubekelaphambili yeqiniso ye-AI kuzibalo. Ngokwenza ilebula ehlukile, i-Epoch AI ivimbela ukweqa amandla e-AI futhi inikeze abacwaningi isithombe esicacile salapho amamodeli asadinga khona ukuqondiswa komuntu. Lokhu kubeka obala kunomthelela ekutheni abatshalizimali, ontanga yezemfundo, kanye nabaqhudelana nabo bahlola kanjani ukuvuthwa kwezinhlelo zokucabanga ze-AI futhi kungase kulolonge imiklamo yebhentshimakhi yesikhathi esizayo ehlose ukujezisa amasu ezinqamuleli.
Ukwehlukanisa ukuzimela nokusebenza okusiziwe kubalulekile ekubekeni okulindelwe okungokoqobo mayelana nekhono le-AI lokwenza ukucabanga kwezibalo kwasekuqaleni ngaphandle kokufaka komuntu. Impumelelo ye-AI eyeqile ingaholela ekuxhasweni kwemali ngendlela engafanele, ukuthunyelwa ngaphambi kwesikhathi, kanye nezinqumo zenqubomgomo ezisekelwe ekuhloleni okungalungile kokuphepha nokwethembeka kwe-AI.
Uhlelo olusha lokulebula luhlinzeka ngemethrikhi eyimbudumbudu kakhudlwana kubacwaningi abahlola amakhono amamodeli, okukhuthaza ukuthuthukiswa kwamasistimu angacabanga ngempela ngaphandle kokwaziswa komuntu. Iphinde inikeze isifanekiso sabanye abadali bebhentshimakhi ukuthi basebenzise ukuhlukaniswa okufanayo, okugqugquzela amazinga embonini yonkana yokubika ukuqhubeka kwe-AI.
Kubatshalizimali kanye nosomaqhinga bezinkampani, ukucaciswa kusiza ukuhlukanisa phakathi kwamamodeli angaxazulula ngokuzimela izinkinga eziyinkimbinkimbi nalawo asathembele kakhulu ekuqondisweni kochwepheshe, okunomthelela ezinqumweni zokuthi zihlukaniseke kuphi izinsiza ukuze kwenziwe ucwaningo olwengeziwe nokuthuthukiswa komkhiqizo.
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
Izibuyekezo ze-Future FrontierMath zizodalula ukuthi isigaba "somuntu + AI" siyakhula nokuthi zingaki izinkinga ezivulekile ezisele ezisuka kokungaxazululiwe ziye ezixazululiwe ngaphansi kwale lebula. Izibukeli kufanele futhi zibuke ukuthi amanye amalebhu e-AI abika kanjani imiphumela yesilinganiso—ukuthi asebenzisa ukulebula okufanayo noma afuna impumelelo yokuzimela—nokuthi umphakathi uyazamukela yini izincazelo ezimisiwe zokusebenza kwe-AI “yokuzimela” kanye “nokusizwa” kwe-AI.
Ukuthi izinkinga ezengeziwe ze-FrontierMath zizohlukaniswa yini kabusha ngaphansi kwelebula elithi “human + AI” njengoba kuvela izixazululo zokusebenzisana ezengeziwe.
Uma amalebhu e-AI ancintisanayo eqala ukushicilela eyawo imiphumela yebhentshimakhi enokulebula okuqhathanisekayo, okungase kuholele ezingeni lemboni ye-de-facto lokubika ukusizwa kuqhathaniswa nezixazululo ezizimele.
Umthelela walokhu kufakwa ilebula emizuliswaneni yoxhaso yesikhathi esizayo yezinkampani ezigxile ekucabangeni kwezibalo okuzimele kuqhathaniswa nalezo ezigcizelela izindlela zomuntu ezingaphakathi kwe-loop.