+− THE DAILY DIFFdev & AI news
SHIP IT

UClaude wafakazela uFermat ezinsukwini ezingu-11. Isinqumo: SHIP IT.

UClaude usebenzise izinsuku ezingu-11 kanye namathokheni ayizigidi eziyisi-6 ebhalini lobuFakazi beLean obunezingu-13-million-line zeFermat's Last Theorem — okokuqala okuhlolwe ngekhompuyutha kusukela ekuqaleni kuze kube sekugcineni — kanti usomathuba obeselokhu ekwenza kube semthethweni kusukela ngo-2024 uthi "akusitsheli lutho" ngokwemathematika futhi ujabule noma kunjalo.

Want to see our stories first on Google? Add us to your preferred sources — it helps us grow.

UClaude usebenzise izinsuku ezingu-11 kanye namathokheni ayizigidi eziyisi-6 ebhalini lobuFakazi beLean obunezingu-13-million-line zeFermat's Last Theorem — okokuqala okuhlolwe ngekhompuyutha kusukela ekuqaleni kuze kube sekugcineni — kanti usomathuba obeselokhu ekwenza kube semthethweni kusukela ngo-2024 uthi "akusitsheli lutho" ngokwemathematika futhi ujabule noma kunjalo. Ngalolo suku olufanayo: Inombolo 1 emhlabeni ku-Go uShin Jin-seo ushaye uKataGo 2-1 ngesandla samatshe amabili. Isinqumo: SHIP IT.

Lokho okuhlanganiswa yile vidiyo

  • UClaude ufaka isandla kuFermat's Last Theorem kuLean 4
  • I-Chromium sandbox RCE (CVE-2026-85046), esetshenziswa kabanzi, umvuzo wama-$1,000
  • UShin Jin-seo ushaye uKataGo ngesandla samatshe amabili

Imibuzo nezimpendulo

Yini uClaude ayifakazela ngempela lapho enza iFermat's Last Theorem ibe semthethweni ku-Lean?

UClaude ubhale ubufakazi be-Lean 4 obuhloliwe ngekhompyutha beFermat's Last Theorem, okokuqala okuvela ekuqaleni kuya ekugcineni. Ama-ejenti e-Anthropic athathe izinsuku ezingu-11 kanye namathokheni aphumayo cishe ayizigidi eziyisi-6, ekhiqiza izigidi eziyi-13 zemigqa ye-Lean kanye nama-theorems aphakathi angu-29,500. I-Lean ihlola isinyathelo ngasinye ngokusebenzisa umshini, futhi ukwakha kuyehluleka ngaphandle kokuthi ubufakazi busekele ezakhiweni ezintathu ezijwayelekile ze-Lean.

www.anthropic.com · github.com

Kuthatha malini ukuphikisa ubufakazi obubhalwe umshini obungako?

Ukuhlola kubiza kakhulu. Ukwakha kusukela ekuqaleni kwathatha amahora amahlanu nesigamu ku-96 cores kanye nama-gigabytes angu-153 e-RAM. UKevin Buzzard uhlanganise i-repository emshinini wama-gigabytes angu-500 owabolekwa yi-Anthropic futhi waqinisekisa ukuthi ihlola. Amagama e-theorem akhiqizwa umshini, ngakho i-repository izichaza njengebhalelwe ukuhlolwa kunokufundwa.

github.com · xenaproject.wordpress.com

Ingabe ubufakazi obusemthethweni obubhalwe i-AI beFermat's Last Theorem bungeza noma yini kumathisiki?

UKevin Buzzard, oholile umzamo wokufomula wabantu kusukela ngo-2024, ubhale ukuthi ngokwezibalo lo msebenzi awusitsheli lutho olubalulekile. Wayesevele eqiniseke ngo-99.9% ukuthi isisho sasiyiqiniso futhi ubufakazi abungezi noma yimuphi umshado omusha. Ujabule noma kunjalo, ngoba kubonisa ukuthi ukuzenzakalela kungakwenza manje. Iposi le-Anthropic lithi i-Lean ibonisa ukunemba ngaphandle kokungabaza.

xenaproject.wordpress.com · www.anthropic.com

Ingabe umuntu angakwazi namanje ukushaya i-AI ye-Go enamandla kakhulu?

Yebo, ngesandla. UShin Jin-seo, inombolo yokuqala emhlabeni ku-Go, washaya uKataGo, injini ye-Go enamandla kakhulu enomthombo ovulekile, imidlalo emibili kweyodwa e-Seoul ngesandla samatshe amabili. Ukunquma kwaba ukunqoba kwamaphuzu angu-11.5 emigudwini engu-221. Uthe walahlekelwa lapho ekopisha izinyathelo ze-AI ekuqaleni futhi wanqoba ngokwakha ibhodi ngesitayela sakhe.

www.kedglobal.com · news.ycombinator.com

Ingabe abathuthukisi nosomathuba kufanele babheke ubufakazi be-Lean kaClaude beFermat njengomphumela wangempela?

Isinqumo se-Daily Diff sasiyi-SHIP IT. I-Lean kernel iyakwamukela ubufakazi futhi u-Kevin Buzzard waqinisekisa ukuthi buhlola. Izibalo zihlala zifana, ngenkathi indlela osomathuba abahlola ngayo ubufakazi ishintshile. Umzamo wabantu u-Buzzard awuholayo waqala ngo-2024, kanti ama-ejenti ka-Claude aqeda ezinsukwini eziyishumi nanye.

www.anthropic.com · xenaproject.wordpress.com

Umbhalo obhaliwe ohunyushwe

Kuhunyushwe kusuka ekulandiseni kokuqala kwesiNgisi. Umsindo namazwibela atholakalayo alawulwa yi-YouTube.

0:00 UFermat wathi ubufakazi bakhe obumangalisayo ngeke buhambisane nomkhawulo, futhi namuhla i-Anthropic ishicilele umkhawulo: izigidi eziyishumi nantathu zemigqa ye-Lean, okuphindwe kahlanu ubukhulu be-Mathlib, kufakazela isisho sonke usomathuba esesivele esikholelwe. Kwakuyishumi nanye eTbilisi lapho i-Anthropic iposa khona, ngakho-ke ngokwemvelo ngangivukile. Izolo i-Google ishiye i-Chrome 152 inezilungiso eziyishumi nambili zokuphepha, enye yazo iyiphutha le-V8 elisetshenziswe kabanzi, futhi yakhokha umbiki i- izinkulungwane zamaRandi, okungaphansi kwesedani esizofika kuyo kamuva.

0:26 Futhi izolo, i-Mullvad ithe ivala i-DNS yayo yomphakathi evikelwe ngekhodi ngoNovemba 2nd futhi ikhokhela i-Quad9 ukuthi yenze lokho esikhundleni sayo, futhi namuhla i-Rust React I-Compiler ibe neyakwawo ku-Vite, ngenkathi i-Hacker News ithola i-IBM Bob, i-ejenti yokubhala ikhodi ye-AI. Ngemva kwalokho uClaude wenza kwaba semthethweni iFermat's Last Theorem, futhi ekhasini elifanayo lokuqala umfundisi omkhulu waseKorea washaya injini ye-Go enamandla kunawo wonke Emhlabeni, ngakho namuhla isintu siphumelele kwababili. Kulolu video: lokho uClaude akufakazela ngempela, lokho akubiza,

0:52 kungani usomathuba osele umsebenzi wakhe kulesi sihloko athi akukho okushintshile futhi ujabule noma kunjalo, nokuthi umuntu washaya kanjani umshini ku-Go. NgoLwesihlanu, Septhemba 4, futhi lena yiThe Daily Diff. IFermat's Last Theorem: azikho izinombolo eziphezulu a, b, c ezanelisa u-a kuya ku-n kanye no-b kuya ku-n kulingana no-c kuya ku-n kunoma yikuphi u-n ongaphezu kuka-2. UFermat wakubamba emkhawulweni cishe ngo-1637 wafa engawubonisi umsebenzi wakhe, okumenza abe unjiniyela wokuqala ukuvala ithikithi ngemisebenzi emshinini wami. Umklomelo ka-1908 wama-mark egolide angu-100,000 waheha iziphambeko ezingu-621

1:25 obufakazi bonyaka wokuqala, kanti u-Andrew Wiles wakuthola ekugcineni ngo-1995, emakhasi angu-129 okwathatha abahlola izinyanga ukukuqinisekisa. Ukufaka isandla kusho ukubhala kabusha lobo buhlola ukuze i-Lean, umsizi wobufakazi, ikwazi ukuhlola isinyathelo ngasinye ngokusebenzisa umshini, kanti u-Kevin Buzzard e-Imperial uhola umuntu umzamo wokwenza lokho kusukela ngo-2024; umdwebo nje unamakhasi angu-86 amakhasi. Umcwaningi we-Anthropic u-Tianyi Peng ukhombe inqwaba yama-ejenti ka-Claude kuwo esikhundleni salokho, endaweni ebizwa nge-Prove2Me egcina i-DAG yesisho se-theorem izenzo ukuze ama-ejenti azi ukuthi yini okufanele ayifakazele ngokulandelayo, ngoba ngaphandle kwayo amaningi okuqala

2:00 alahlekelwa ukulandela ukuthi ubani ofakazela ini, okuyilokho okwenzeka lapho i- ungqimba lwakho lokuhlela luyi-regex enesabelomali sokumaketha. Ezinsukwini eziyishumi nanye kamuva i-root node yafunda i-PROVED: izigidi eziyishumi nantathu zemigqa ye-Lean, izigidi ezingama-29,500 zamathiyori aphakathi, cishe izigidi eziyisithupha zamathokheni aphumayo ku- imodeli yangaphakathi efana kakhulu ne-Claude Fable 5.1. Ukuhlanganisa kuyehluleka ngaphandle kokuthi ubufakazi busekele ezakhiweni ezintathu ezijwayelekile ze-Lean: cha uxolo, akukho ukunquma kwasekuqaleni, akukho ukukopela. Ukukuhlola akushibhile nakancane: i-

2:29 ukwakha kusukela ekuqaleni kwathatha amahora amahlanu nesigamu ku-96 cores kanye nama-gigabytes angu-153 e-RAM, futhi amagama e-theorem akhiqizwa umshini, ngakho i-repo izichaza njengebhalelwe ukuhlolwa kunokufundwa, okuyiyo futhi indlela engingachaza ngayo i-enterprise Java. Manje isiphikisana. Iposi le-Anthropic lithi i-Lean ikhombisa ukunemba ngaphandle kokungabaza. UKevin Buzzard, umuntu owashaywa yiyo, uhlanganise i-repo emshinini wama-gigabytes angu-500 i-Anthropic yamboleka yena, yaqinisekisa ukuthi ihlola, bese ibhala,

2:56 ngqo, ngokwemathematika lo msebenzi awusitsheli lutho. Wayesevele eqiniseke ngo-99.9% ukuthi isisho sasiyiqiniso, futhi ubufakazi abungezi noma yimuphi umshado omusha; lokho okubonisa ukuthi ukufomula okuzenzakalelayo kungakwenza manje, futhi leyo ngxenye ujabule kakhulu ngayo. Wanikezwa isigidi esisodwa samaphawundi eminyakeni emihlanu; i-Anthropic yathatha izinsuku eziyishumi nanye, futhi ukubala kwephepha lomhlaziyi kubeka izigidi eziyisithupha zamathokheni aphumayo ngentengo yohlu cishe ama-dollar angama-300,000, ngakho umshini wawushibhile, ngaphandle kokuthi ubale ukuqeqesha umshini, okungekho muntu okwenza.

3:24 Imininingwane ehamba phambili: i-imeyili yafika ngesikhathi esephathi yomculo e-Wales ne- ibha elilodwa le-4G, elivela egameni ayengakaze alizwe, ngakho walibala njengomuntu obhadla futhi walifunda ngemuva kwesonto, okuyiyo impendulo efanele kunoma yimuphi umugqa wesihloko oqukethe ukufomula okuphelele. Khonamanjalo, abantu bathole okukodwa emuva. UShin Jin-seo, inombolo yokuqala emhlabeni ku-Go, washaya uKataGo, injini ye-Go enamandla kakhulu enomthombo ovulekile, imidlalo emibili kweyodwa e-Seoul ngesandla samatshe amabili, cishe igebe eliphakathi kochwepheshe abaphezulu kanye nochwepheshe abasafufusa.

3:50 Ukunquma kwakuwukunqoba kwamaphuzu angu-11.5 emigudwini engu-221, kubambe i- Izinga lokunqoba elingu-99% kusukela phakathi komdlalo, futhi wathatha ekhaya izigidi ezingu-250 wawina, cishe ama-dollar angu-170,000, kanye ne-Genesis G90, ngakho umvuzo wokunqoba i- I-AI enamandla kakhulu iphindwe izikhathi ezingu-170 umvuzo we-Google wokuphuma e-Chrome sandbox. Incazelo yakhe: ekuqaleni wakopisha izinyathelo ze-AI walahlekelwa; wanqoba nge- ukwakha ibhodi ngesitayela sakhe, okuyisihloko esiwusizo kakhulu nge-AI engasoze ngisizwile kulo nyaka, futhi kwavela kumdlalo webhodi. Imigqa emibili eyengeziwe ku-diff.

4:22 I-Rust React Compiler evela ku-oxc manje isingumuntu omdabu ku-Vite ngemuva kweflegi elilodwa; i- Ikhodi yamafayela angu-1,036 yahlala kusukela kumasekhondi angu-14.3 kuya ku-0.81 ku- isinyathelo sokuhlanganisa, ikakhulukazi ngokususa i-Babel ku- package.json, okuyiyo futhi indlela yami yokunakekela isikhumba. Futhi i-IBM yethule u-Bob, umngane wokubhala ikhodi we-AI okubingelela nge-Sawubona, Ngingu-Bob, ukhulisa ama-subagents, ubeka ikhodi ye-mainframe yesimanje, futhi ihambisa umkhiqizo wokuhlaziya obizwa nge-Bobalytics, ngakho-ke kwenye indawo ibhange lijabule kakhulu futhi akekho owayefunda ilayisense.

4:51 Lokho kungumkhawulo omkhulu kakhulu wangolwesihlanu olulodwa; uma ungathanda ukufunda lokhu kunokungizwa ngikushoyo, i-diff ifika ebhokisini lakho le-imeyili njalo ekuseni — mahhala ku-the daily diff dot dev, isixhumanisi ngezansi. Ngakho, isinqumo sanamuhla: SHIP IT. I-kernel ithi yebo, u-Buzzard uthi yebo, umshado awuzange ushintshe, kodwa indlela esihlole ngayo umshado isanda kushintsha. Leyo yi-diff yanamuhla. Ngingu-Niko ovela ku-Axrisi.

5:09 Hlanganisa ngokuzibophezela.

Imithombo

  1. Anthropic — Formalizing Fermat's Last Theoremwww.anthropic.com
  2. The proof (Lean 4, Apache-2.0)github.com
  3. Kevin Buzzard — FLT: Anthropic has beaten me to itxenaproject.wordpress.com
  4. HN threadnews.ycombinator.com
  5. KED Global — Shin defeats KataGowww.kedglobal.com
  6. HNnews.ycombinator.com
  7. Chrome 152 release notes (CVE-2026-85046)chromereleases.googleblog.com
  8. NVDnvd.nist.gov
  9. Mullvad — shutting down public encrypted DNSmullvad.net
  10. Rust React Compiler native in Viteblog.master.dev
  11. IBM Bobbob.ibm.com

Niko from Axrisi

Nikoloz Turazashvili · Vexrail · Axrisi

LinkedIn · dev.to · GitHub · X · Substack

Amavidiyo ahlobene