I-OpenAI ikhiphe imiphumela yokuqala ye- I-Astra, imodeli yabo elandelayo yobuhlakani bokwenziwaLe nkampani ithuthuke ezinkingeni eziyishumi ezivulekile zezibalo kanye nesayensi yamakhompyutha yethiyori. Ngokusho kwenkampani, ezinye zalezi zinkinga bezilokhu zingaxazululwa iminyaka engaphezu kweshumi ngaphandle kwentuthuko ebalulekile. Lesi simemezelo, esanyatheliswa ngo-Agasti 1, 2026, sidale injabulo enkulu emphakathini wesayensi, yize abacwaningi abaningi benxusa ukuba kuqashelwe kuze kube yilapho imiphumela isibuyekezwe ochwepheshe abazimele.
Okuhlukanisa lesi sikhangiso kwezinye ezifanayo ukuthi I-OpenAI ayishicilelanga imiphumela kuphela, kodwa futhi nemiboniso esemthethweni ohlelweni lokuqinisekisa lwe-Lean. Lokhu kuvumela noma yimuphi usosayensi wezibalo ukuthi ahlole ngokuzenzakalelayo ukunemba kwezimpikiswano. Ngaphezu kwalokho, inkampani inemininingwane yenqubo yokuthola, kufaka phakathi amakhasi angu-249 ezimpikiswano namakhasi angu-62 amanothi anengqondo, okunikeza ukucaca okungavamile kulolu hlobo lokuxhumana.
Izinkinga eziyishumi nokubaluleka kwazo
Izinkinga ezixazululwe yi-Astra zihlanganisa izindawo ezahlukahlukene kakhulu zezibalo zanamuhla. Phakathi kwazo kukhona: ukuphikiswa kokuqagela kukaConnes okuqinileLo mphumela unomthelela omkhulu kuma-algebra opharetha, futhi ukwakhiwa okucacile kwamaqembu angewona ama-Sophic kuyinkinga eyisisekelo kumbono weqembu. Kutholakale nemingcele emisha ephezulu yokupakisha ama-sphere ngobukhulu obuphezulu, inkinga ngokusetshenziswa okuqondile kumbono wolwazi kanye namakhodi okulungisa amaphutha.
Emkhakheni we-combinatorics, i-Astra ixazulule izinkinga ezintathu ezibangelwe yisazi sezibalo saseHungary uPaul Erdős: izinkinga ezingu-146, 180, kanye no-183. Eyokugcina yalezi, ehlobene nezinombolo zikaRamsey ezinemibala eminingi, ichazwe uchwepheshe uThomas Bloom ngokuthi "imangalisa kakhulu ohlwini lonke." Ezinye intuthuko zifaka phakathi imingcele emisha ephansi yamasekethe ezibalo, i-parallel repetition theorem yemidlalo ye-quantum, kanye nemiphumela ebulukhunini bokusondela kwenkinga ye-neastern-vector, ehlobene ne-post-quantum cryptography.
Izindleko kanye nendlela yokwenza izinto
I-OpenAI iveze ukuthi Izindleko eziphelele zokuthola lezi zixazululo bezicishe zibe ngu-$2.000Ngokusho kwamazinga e-API kaSol, lesi sibalo asifaki ukuqeqeshwa kwemodeli noma umsebenzi wabacwaningi abangabantu. Inqubo ihlanganise ukukhiqizwa kwempikiswano yi-Astra kanye nokufaka kochwepheshe okwasiza ukuguqula lezo zimpikiswano zibe yimibhalo ecacile, futhi kamuva, zaba izitifiketi ze-Lean eziqinisekiswa ngokuzenzakalelayo.
Ukwenza kube semthethweni kubalulekile ku-Lean: kuvumela noma ubani onokufinyelela kusofthiwe ukuqinisekisa isinyathelo ngesinyathelo ukuthi ukuboniswa kulungile ngokunengqondo. Kodwa-ke, Izazi zezibalo kusadingeka ziqinisekise ukuthi izincazelo kanye nokuqagela okusetshenzisiwe kuyahambisana nenkinga yokuqala.Lokhu kuyinto umshini ongenakukwazi ukuyenza wodwa. I-OpenAI ithathe umthwalo wokulungisa umsebenzi, kodwa ayikahlonzi noma yiliphi iqembu elizimele elihlole yonke imiphumela eyishumi esidlangalaleni.
Ukuqhathaniswa ne-Deep Blue kanye nokuqapha okudingekayo
Imithombo yezindaba eminingana iqhathanise lesi sigameko nokunqoba kukaDeep Blue uGarry Kasparov ngo-1997, kodwa ochwepheshe baveza ukuthi Izibalo ziyinkimbinkimbi kakhulu kune-chessEmdlalweni, umphumela uba ngokushesha; kwizibalo, ubufakazi bungadinga izinyanga zokubuyekezwa kontanga ngaphambi kokuba bamukelwe. Ngaphezu kwalokho, i-Astra ayitholakali emphakathini, ngakho abacwaningi bangaphandle abakwazi ukuphinda inqubo noma ukunquma izinga lokwehluleka kwemodeli.
I-OpenAI ibilokhu ihlangabezana nakho ngaphambilini lapho imiphumela ebonakala ilungile yabonakala ingalungile ngemva kokubuyekezwa ochwepheshe. NgoFebhuwari 2026, inkampani yathumela izixazululo zenselele ye-First Proof futhi kamuva yavuma ukuthi enye yazo yayingalungile. Ngakho-ke, Umphakathi wesayensi uphikelela ngokuthi lezi ntuthuko eziyishumi kufanele ziphathwe njengezithembisayo, kodwa hhayi njengeziqinisekile.kuze kube yilapho ziqinisekiswa ngokuzimela.
EYurophu, abacwaningi abavela emanyuvesi afana nePolytechnic University of Madrid kanye neYunivesithi yaseCambridge baveze intshisekelo yokuhlaziya izitifiketi zeLean ezishicilelwe yi-OpenAI. Ukuthi kungenzeka ukuthi ubuhlakani bokwenziwa busebenzisane ekwakhiweni kolwazi olusha lwezibalo kuvula ingxoxo mayelana nendima yososayensi abangabantu esikhathini esizayo, yize inkampani igcizelela ukuthi ukuqinisekiswa kanye nokuhlanganiswa kwemiphumela kuzoqhubeka kuncike kochwepheshe iminyaka eminingi ezayo.
Noma kunjalo, isimemezelo sika-Astra sibonisa ushintsho olukhulu: ubuhlakani bokwenziwa abusekho ekuxazululeni izinkinga ngezimpendulo ezaziwayo, kodwa buqala ukuphakamisa izimpikiswano zokuqala mayelana nemibuzo ehamba phambili kolwazi. "Umzuzu oluhlaza okwesibhakabhaka ojulile" weqiniso kwizibalo ungase ungafiki nomphumela owodwa omangalisayo, kodwa kancane kancane, njengoba la mamodeli eba abahlanganyeli abakwazi ukuhlola izinkulungwane zezindlela nokwenza imiphumela ibe semthethweni okungekho mcwaningi owake wayihlola ngaphambili.
