Subira ku makuru
Guhanga udushyaAI Understanding ibisobanuro

Ubushakashatsi bwa AI bufashijwe Kwiga Buzuza Imipaka Kumwanya muremure

Ubushakashatsi bwakozwe buvuga ko sisitemu y’ubushakashatsi bwa AI yafashije kunoza imipaka kuri Grothendieck ihoraho, mu gihe abanditsi bashimangira ko abashakashatsi b’abantu bahisemo pivot y'ingenzi kandi bagenzura ubwigenge gusa ibisubizo byo ku rwego rwa theorem.

6 min readRead the primary source
Inyandiko y'ibanzeInkomoko yanditse
Umwanditsi
Long-horizon AI mathematics case study on arXiv
Ihuza ry'inkomoko
arxiv.orghttps://arxiv.org/abs/2608.11195
Ubwoko bw'inkomoko
Inyandiko y'ibanze - itangazo ryemewe, impapuro, dosiye, cyangwa urupapuro rwambere-dusoma mu buryo butaziguye.
ImirongoSobanukirwa ibi mumasegonda 60

Tangira hano

Amagambo y'ingenzi

API (Imigaragarire ya Porogaramu)
Inzira yuburyo bwa sisitemu imwe yohereza ibyifuzo no kwakira ibisubizo bivuye murindi sisitemu.
Kwibuka (Memory Memory)
Imiterere yabitswe umukozi wa AI akoresha intambwe cyangwa amasomo kugirango atezimbere.
Ibipimo
Ikizamini gisanzwe cyangwa dataset ikoreshwa mugupima no kugereranya imikorere yicyitegererezo.
IsuzumeIkibazo cya AI

Byagenze bite

Ubushakashatsi bushya bwa arXiv busobanura uburyo sisitemu y’ubushakashatsi bwa AI yafashije abahanga mu mibare kunoza imipaka izwi kuri Grothendieck ihoraho, ikibazo gifunguye cyatangiye mu 1953. Uru rupapuro rugaragaza ibyavuye mu mibare ndetse n’ibisobanuro birambuye byerekana aho sisitemu yakoraga neza, aho abantu bagombaga kuyiyobora, kandi ikavuga ko ikomeza kugenzurwa n’imashini aho kugenzurwa n’abantu.

Grothendieck ihora ipima itandukaniro riri hagati yikibazo gikomeye cyo guhuza ibitekerezo hamwe no kuruhuka gukabije. Abanditsi bavuga intera nshya ifite imipaka yo hepfo ya 6pi / 11, hafi 1.7135, n’umupaka wo hejuru wa 1.7818. Impapuro zumubare wimpapuro zitanga ibimenyetso. Ibisubizo byo hasi-ibisubizo biragaragara kuko ntabwo byubaka urugero rukomeye; ahubwo ikuramo inzitizi ikoreshwa murwego rwo kuzenguruka.

Sisitemu yubushakashatsi yahujije uburyo bwo gutekereza hamwe na code ya agent. Uru rupapuro ruvuga ko kwiruka byakoresheje GPT-5.5-Pro ​​hanyuma nyuma GPT-5.6-Sol yo gutekereza, Claude Code hamwe na Opus hanyuma nyuma ya Fable 5 kugirango ikorwe, na node ya GPU ine yo gushakisha imibare. Amasomo yatwaye ubudahwema binyuze muri dosiye, inyandiko-mvugo, inyandiko zigerageza, hamwe nincamake yanditseho ibirego nkuko byemejwe, bishyigikiwe numubare, ibitekerezo, cyangwa heuriste aho gushingira kumurongo umwe udahagarara.

Iri rushanwa ryakubiyemo amasomo agera kuri 240 y’ubushakashatsi kuva ku ya 16 Kamena kugeza ku ya 24 Nyakanga, hahamagarwa 2.091 yo gutekereza-hamwe n’ibimenyetso bigera kuri miliyoni 152 ku giciro cya $ 5.400. Amatariki agera kuri 40 yubuyobozi bwabantu yayoboye akazi. Mugihe ubushakashatsi bwo hejuru bwashizwe hejuru, ababikora bamenye ko kunanirwa inshuro nyinshi bishobora kwerekana inzitizi rusange hanyuma berekeza sisitemu yerekeza kumpaka zo hasi. Uru rupapuro rusobanura ko guhindura icyerekezo cyubushakashatsi nkigikorwa cyabantu, ntabwo ari icyemezo cyigenga.

Sisitemu yabyaye reframe yo hagati hamwe nibimenyetso byuzuye kumurongo wa 6pi / 11 wo hepfo, abanditsi bavugurura hanyuma barabigenzura. Yabyaye kandi umubare munini w’abakandida bo hejuru n’abato batsinze protocole yimbere, ariko abanditsi ntibigeze bigenzura ibyo byemezo kandi ntibabivuga nka theorem. Urupapuro rero rutandukanya ibisubizo byimibare byagenzuwe na sisitemu irenze urugero cyangwa imashini yageragejwe.

Ibisobanuro birambuye: Long-horizon AI mathematics case study on arXiv ↗

Impamvu ari ngombwa

Ubushakashatsi butanga ibimenyetso bifatika bidasanzwe byerekeranye na AI ikoreshwa muri gahunda yubushakashatsi aho kuba umurimo umwe. Icyagaragaye cyane ni igabana ry'imirimo: sisitemu yari ikomeye mu bikorwa bya tekiniki, mu gihe abashakashatsi b'abantu bakomeje kuba inshingano zo gushyiraho gahunda, guca imanza, no kugenzura bwa nyuma.

Ubushakashatsi burebure bwa horizon buragoye kuri sisitemu ya AI kuko intego irashobora guhinduka nkuko inzira zananiranye zegeranya. Umufatanyabikorwa w'ingirakamaro agomba kugumana ibyageragejwe, gutandukanya iherezo ryapfuye nigitekerezo kidakemutse, no guhitamo igihe ikibazo gishya gifite agaciro kuruta ikindi kintu cyiza. Ubwanditsi bwa dosiye ishingiye kububiko hamwe nibirango bisaba ni ukugerageza gukora ubwo bushakashatsi bugaragara kandi bugenzurwa aho kwemerera igisubizo cyiza guhagarara imbere.

Ubuvumbuzi buke bugaragaza impamvu urubanza rwabantu rugifite akamaro nubwo umukozi ashobora gukora imibare. Abakoresha babonye ko kugerageza kubaka byinshi byananiranye muburyo bumwe kandi batanga pivot yibitekerezo: garagaza inzitizi ubwayo. Sisitemu noneho yafashaga kwemeza no kwemeza impaka. Urwo ruhererekane rwegereye ubushakashatsi bugenzurwa kuruta imibare yigenga yimashini yigenga, kandi itandukaniro rifite akamaro mugihe dusobanura ibyo ibimenyetso bifatika.

Igenzura ryigenga ni irembo ryo kurekura ibisubizo. Ubushakashatsi bwakozwe buvuga ko imipaka yo hepfo yagenzuwe n’abanditsi kandi ko ibimenyetso byuzuye bigaragara mu mpapuro za mugenzi. Ntabwo bivuze ko buri mubare wakozwe mugihe cyo kwiruka ari byo. Kugumya kurwego rwo hejuru rwibisabwa, abakandida bapimwe imashini, hamwe na hypotheses bitandukanye biha abasomyi uburyo bwo gusuzuma umusanzu batiriwe bemera sisitemu yimbere ya AI nkibimenyetso byimibare.

Ku mashyirahamwe yubushakashatsi, isomo rifatika rirakora. Sisitemu ya AI irashobora gushakisha ubuvanganzo, kwandika kode, gukora igerageza, kubika ibigeragezo byananiranye, no gutanga impinduka, ariko urujya n'uruza rw'akazi rukeneye ibimenyetso, kubara kubyara, kugenzura abantu, ndetse nuburyo bwo kongera kubaka impamvu itsinda ryahinduye icyerekezo. Ibiciro byatangajwe hamwe na compte nabyo birerekana neza ko ubushobozi burebure bwa horizon atari ikibazo cyubwenge bwikitegererezo gusa; biterwa na scafolding, igihe, hamwe nubugenzuzi burambye.

Interactive Mechanism

Uburyo bukoreshwa: Uburyo bukora

Shakisha ikoranabuhanga ryihishe inyuma yiri terambere.

Thinking Budget (Test-Time Tokens):1,024 tokens
Complex Accuracy79%Math & Code Logic
Latency3.2sTime to first full output
Inference Cost$0.0092Per query estimated
Reasoning StyleStep VerificationInternal chain depth
Active Thinking Trace:
1Deconstruct user problem into formal constraints
2Propose candidate hypotheses & step-by-step calculation
3Self-correction: Backtrack and refute subtle edge cases
4Exhaustive consistency check & final output synthesis
Core takeaway: Test-time compute fundamentally changes AI economics. Instead of only scaling during pre-training, giving reasoning models more tokens at inference time allows them to systematically solve PhD-level STEM problems.
Kugenzura Ibitekerezo Byagenzuwe+10 Points
AI Agents Quiz

What most distinguishes an AI agent from a basic chatbot?

Ibyo kureba

Ikibazo gikurikiraho ni ukumenya niba ubu buryo bwubufatanye busanzwe burenze ikibazo kimwe nitsinda, kandi niba abashakashatsi bigenga bashobora kubyara ibisubizo byemejwe mugihe bapima ikiguzi nubwizerwe bwa buri musanzu wabantu na AI.

Kwigana bigomba kugerageza izindi mibare, ubwoko bwibibazo, hamwe na sisitemu yubushakashatsi hamwe nibikoresho bitandukanye byo kwibuka. Ikibazo cya Grothendieck kimaze kuzana ibintu byinshi byubatswe n'abantu, inoti zegeranijwe, imashini zemeza ibyemezo, hamwe nicyerekezo cyabakandida. Iyo miterere ibanza yafashaga gukora, bityo ubushakashatsi buzaza bugomba kwerekana umubare wubushakashatsi bwatanzwe mbere nuburyo ibisubizo bihinduka mugihe sisitemu igomba gusobanura ikibazo ubwacyo.

Abashakashatsi bagomba kandi kugereranya ishyirwa mubikorwa rya tekiniki hamwe nubushakashatsi bwakozwe hakoreshejwe ingamba ziteganijwe. Ibipimo byingirakamaro bishobora kuba bikubiyemo ubwiza bwicyerekezo cyatoranijwe, igihe cyo kureka inzira yananiwe, igipimo cyibisabwa bidashyigikiwe, umubare wabantu bitabira, hamwe nigice cyibisubizo bikomeza kugenzurwa byigenga. Ubushakashatsi bwakozwe neza burashobora kwerekana uko byagenze, ariko kugereranya kugenzurwa birakenewe kugirango ugereranye mugihe umufatanyabikorwa wa AI atezimbere inzira aho kongeramo urundi rwego rwo gusuzuma.

Imipaka iri hagati yo kugenzura imashini no kugenzura abantu ikwiye kwitabwaho. Intera yimibare, abafasha berekana ibimenyetso, ibizamini, hamwe nubugenzuzi bwa adversarial birashobora gufata amakosa menshi, nyamara icyemezo gishobora gukomeza gushiramo intego itari yo cyangwa biterwa nibitekerezo bitigeze bivuguruzanya. Gukurikirana akazi bigomba gutangaza ibihangano byuzuye, gusubiramo imipaka ikomeye itagenzuwe, no gukora ibisubizo byatsinzwe cyangwa byakuweho nkibigaragara neza.

Hanyuma, verisiyo yicyitegererezo, ibisobanuro, amabwiriza yo kuyobora, kode, kubara, hamwe ninyandiko-mvugo bigomba kubikwa aho uruhushya rutangwa. Impapuro ziriho zifite agaciro igice kuko zitanga ibisabwa kandi zigasobanura intege nke za sisitemu, harimo ubushakashatsi bworoshye-leta ihagarariye hamwe nubwigenge buke mu guca imanza. Kugeza igihe andi makipe azabyara icyitegererezo, ibi nibimenyetso bikomeye byerekana iterambere rya mibare ifashwa na AI, ntabwo ari gihamya ko sisitemu ya AI ishobora kwigenga gukora ubushakashatsi bwuzuye.

Ibijyanye nuyobora & ibibazo

Abakozi ba AIModeri ya AI YasobanuweAmahugurwa ya AIIsuzuma rya LLMGerageza ibyo uzi - gerageza ikibazo cya AI kubuntuReba ijambo AI mumagambo yacuKurikiza icyerekezo cya AI cyo kurekura
Basanze ari ingirakamaro?