#alphaproof — Public Fediverse posts

https://winbuzzer.com/2026/05/26/google-deepmind-says-alphaproof-nexus-is-still-not-agi-xcxwbn/

DeepMind’s AlphaProof Nexus solves nine Erdős problems using Lean-checked proofs, signaling a new phase in AI math after OpenAI’s geometry claim.

#AI #AlphaProofNexus #AlphaProof #GoogleDeepMind #Google #GoogleGemini #AIResearch #Erdős #Math

https://winbuzzer.com/2026/05/26/google-deepmind-says-alphaproof-nexus-is-still-not-agi-xcxwbn/

DeepMind’s AlphaProof Nexus solves nine Erdős problems using Lean-checked proofs, signaling a new phase in AI math after OpenAI’s geometry claim.

#AI #AlphaProofNexus #AlphaProof #GoogleDeepMind #Google #GoogleGemini #AIResearch #Erdős #Math

https://winbuzzer.com/2026/05/26/google-deepmind-says-alphaproof-nexus-is-still-not-agi-xcxwbn/

DeepMind’s AlphaProof Nexus solves nine Erdős problems using Lean-checked proofs, signaling a new phase in AI math after OpenAI’s geometry claim.

#AI #AlphaProofNexus #AlphaProof #GoogleDeepMind #Google #GoogleGemini #AIResearch #Erdős #Math

https://winbuzzer.com/2026/05/26/google-deepmind-says-alphaproof-nexus-is-still-not-agi-xcxwbn/

DeepMind’s AlphaProof Nexus solves nine Erdős problems using Lean-checked proofs, signaling a new phase in AI math after OpenAI’s geometry claim.

#AI #AlphaProofNexus #AlphaProof #GoogleDeepMind #Google #GoogleGemini #AIResearch #Erdős #Math

https://winbuzzer.com/2026/05/26/google-deepmind-says-alphaproof-nexus-is-still-not-agi-xcxwbn/

DeepMind’s AlphaProof Nexus solves nine Erdős problems using Lean-checked proofs, signaling a new phase in AI math after OpenAI’s geometry claim.

#AI #AlphaProofNexus #AlphaProof #GoogleDeepMind #Google #GoogleGemini #AIResearch #Erdős #Math

Readings shared April 4, 2026. https://jaalonso.github.io/vestigium/posts/2026/04/04-readings_shared_04-04-26 #AI #AI4Math #ATP #Agda #AlphaProof #Autoformalization #CategoryTheory #CoqProver #FunctionalProgramming #ITP #IsabelleHOL #LLMs #LambdaCalculus #LeanProver #Lisp #Logic #LogicProgramming #LLMs #Math #Physics #Programming #Prolog #Racket #RocqProver #Vampire

Readings shared April 4, 2026. https://jaalonso.github.io/vestigium/posts/2026/04/04-readings_shared_04-04-26 #AI #AI4Math #ATP #Agda #AlphaProof #Autoformalization #CategoryTheory #CoqProver #FunctionalProgramming #ITP #IsabelleHOL #LLMs #LambdaCalculus #LeanProver #Lisp #Logic #LogicProgramming #LLMs #Math #Physics #Programming #Prolog #Racket #RocqProver #Vampire

Readings shared April 4, 2026. https://jaalonso.github.io/vestigium/posts/2026/04/04-readings_shared_04-04-26 #AI #AI4Math #ATP #Agda #AlphaProof #Autoformalization #CategoryTheory #CoqProver #FunctionalProgramming #ITP #IsabelleHOL #LLMs #LambdaCalculus #LeanProver #Lisp #Logic #LogicProgramming #LLMs #Math #Physics #Programming #Prolog #Racket #RocqProver #Vampire

Readings shared April 4, 2026. https://jaalonso.github.io/vestigium/posts/2026/04/04-readings_shared_04-04-26 #AI #AI4Math #ATP #Agda #AlphaProof #Autoformalization #CategoryTheory #CoqProver #FunctionalProgramming #ITP #IsabelleHOL #LLMs #LambdaCalculus #LeanProver #Lisp #Logic #LogicProgramming #LLMs #Math #Physics #Programming #Prolog #Racket #RocqProver #Vampire

Readings shared April 4, 2026. https://jaalonso.github.io/vestigium/posts/2026/04/04-readings_shared_04-04-26 #AI #AI4Math #ATP #Agda #AlphaProof #Autoformalization #CategoryTheory #CoqProver #FunctionalProgramming #ITP #IsabelleHOL #LLMs #LambdaCalculus #LeanProver #Lisp #Logic #LogicProgramming #LLMs #Math #Physics #Programming #Prolog #Racket #RocqProver #Vampire

Reseña de «How we achieved an IMO medal, one year before any other AI system». https://jaalonso.github.io/vestigium/posts/2025/11/14-how-we-achieved-an-imo-medal-one-year-before-any-other-ai-system/ #AI #Math #ITP #LeanProver #AlphaProof

Readings shared November 23, 2025. https://jaalonso.github.io/vestigium/posts/2025/11/24-readings_shared_11-23-25 #AI #Agda #AlphaProof #FunctionalProgramming #ITP #IsabelleHOL #LLMs #LeanProver #Math #OCaml #Rocq

Reseña de «DeepMind’s latest: An AI for handling mathematical proofs». https://jaalonso.github.io/vestigium/posts/2025/11/21-deepminds-latest-an-ai-for-handling-mathematical-proofs/ #AI #Math #ITP #LeanProver #AlphaProof

DeepMind’s latest: An AI for handling mathematical proofs. ~ Jacek Krywko. https://arstechnica.com/ai/2025/11/deepminds-latest-an-ai-for-handling-mathematical-proofs/ #AI #Math #LLMs #ITP #LeanProver #AlphaProof

AlphaProof od DeepMind: AI zdobyła srebrny medal na Olimpiadzie Matematycznej. Ma to swoją cenę

Komputery są świetne w liczeniu, ale słabe w rozumowaniu. Teraz zespół Google DeepMind ogłosił przełom: AlphaProof, nowy system AI, dorównał srebrnym medalistom Międzynarodowej Olimpiady Matematycznej (IMO) 2024.

Twór DeepMind osiągnął wynik 28 punktów, stając się siódmym podmiotem (obok sześciu ludzi), który rozwiązał najtrudniejsze zadanie.

Jak zauważa ArsTechnica, to ogromny sukces. Do tej pory modele AI nie radziły sobie z dowodami matematycznymi, ponieważ polegały na statystycznym przewidywaniu, co „brzmi” poprawnie, a nie na zrozumieniu struktury matematyki.

TTRL: uczenie się jak człowiek

DeepMind wykorzystało architekturę znaną z AlphaZero (tej od gier: go, szachy), ale dodało trzeci, unikalny element: Test-Time Reinforcement Learning (TTRL). Ten komponent naśladuje podejście człowieka do trudnych problemów.

Kiedy AlphaProof nie potrafi rozwiązać zadania, tworzy setki jego wariacji – uproszczonych, uogólnionych lub luźno powiązanych. Następnie uczy się, próbując rozwiązać te łatwiejsze wersje, aby zdobyć praktykę i nabyć „praktyczne doświadczenie” w trakcie trwania zadania.

Cena srebrnego medalu: dni i setki TPU

Ten sukces ma jednak gigantyczną cenę, co jest kluczowym elementem krytycznej oceny. Po pierwsze, czas. Ludzie uczestniczący w Międzynarodowej Olimpiadzie Matematycznej mieli na rozwiązanie sześciu problemów dwie sesje po cztery i pół godziny. AlphaProof zmagał się z problemami przez… kilka dni, zużywając jednocześnie wiele jednostek TPU (Tensor Processing Unit).

W efekcie cały system potrzebował setek TPU-dni na problem. Jak przyznaje DeepMind, wymagania obliczeniowe są „najprawdopodobniej zbyt kosztowne dla większości grup badawczych”.

Jakby tego było mało AlphaProof nie działał całkowicie autonomicznie. Potrzebował ludzi do przetłumaczenia problemów na formalny język Lean oraz musiał wywołać drugą, wyspecjalizowaną AI (AlphaGeometry 2) do rozwiązania problemu z geometrią.

Podsumowując, AlphaProof to dowód na to, że AI osiągnęło poziom rozumienia logiki, ale do zastąpienia ludzkiego matematyka brakuje mu jeszcze… szybkości, elegancji i pieniędzy. Zatem, gdy ktoś powie, że AI jest dużo szybsza od człowieka, to niniejszy przypadek jest dowodem, że wciąż mamy przewagę.

AWS rzuca wyzwanie Copilotowi. Kiro już dostępne, a startupy dostaną je za darmo

#aiWMatematyce #alphaproof #deepmind #lean #miedzynarodowaOlimpiadaMatematyczna #nature #news #ttrl

Readings shared November 14, 2025. https://jaalonso.github.io/vestigium/posts/2025/11/15-readings_shared_11-14-25 #AI #Agda #AlphaProof #CoqProver #FunctionalProgramming #Haskell #ITP #LeanProver #Math #OCaml #Rocq

How we achieved an IMO medal, one year before any other AI system. ~ Tom Zahavy. https://www.tomzahavy.com/post/how-we-achieved-an-imo-medal-one-year-before-everyone-else #AI #Math #ITP #LeanProver #AlphaProof

Reseña de «Olympiad-level formal mathematical reasoning with reinforcement learning». https://jaalonso.github.io/vestigium/posts/2025/11/14-olympiad-level-formal-mathematical-reasoning-with-reinforcement-learning/ #AI #Math #ITP #LeanProver #AlphaProof

Readings shared October 3, 2025. https://jaalonso.github.io/vestigium/posts/2025/10/04-readings_shared_10-03-25 #AI #ATP #AlphaProof #FunctionalProgramming #Haskell #ITP #LeanProver #Math #Mathlib

Readings shared October 3, 2025. https://jaalonso.github.io/vestigium/posts/2025/10/04-readings_shared_10-03-25 #AI #ATP #AlphaProof #FunctionalProgramming #Haskell #ITP #LeanProver #Math #Mathlib

Readings shared October 3, 2025. https://jaalonso.github.io/vestigium/posts/2025/10/04-readings_shared_10-03-25 #AI #ATP #AlphaProof #FunctionalProgramming #Haskell #ITP #LeanProver #Math #Mathlib

Readings shared October 3, 2025. https://jaalonso.github.io/vestigium/posts/2025/10/04-readings_shared_10-03-25 #AI #ATP #AlphaProof #FunctionalProgramming #Haskell #ITP #LeanProver #Math #Mathlib

Readings shared October 3, 2025. https://jaalonso.github.io/vestigium/posts/2025/10/04-readings_shared_10-03-25 #AI #ATP #AlphaProof #FunctionalProgramming #Haskell #ITP #LeanProver #Math #Mathlib

Mathematics in the age of automated proofs: AlphaProof (From the lab into your hands). ~ Thomas Hubert. https://youtu.be/uhwfTOUIeiw #AI #Math #ITP #LeanProver #AlphaProof

Yapay zeka, büyük veri setlerini analiz etmekten karmaşık simülasyonlar oluşturmaya kadar, hesaplamalı matematiğin ağır işlerini üstlenerek büyük bir güç haline geldi. Şu an itibarıyla, YZ genellikle kendisine verilen ispatları adım adım doğrulayabiliyor; ancak nihai kararı ve en yaratıcı kanıt yöntemlerinin keşfini hâlâ insan zekasına bırakıyor.
Peki ya bu denge değişirse?
#YapayZeka #Matematik #AlphaProof #Lean #mathlib #DeepMind #YaşamOyunu
https://youtu.be/ZERkY5kExjI
https://monologblg.com/matematik-sadece-insana-mi-ait/

Yapay zeka, büyük veri setlerini analiz etmekten karmaşık simülasyonlar oluşturmaya kadar, hesaplamalı matematiğin ağır işlerini üstlenerek büyük bir güç haline geldi. Şu an itibarıyla, YZ genellikle kendisine verilen ispatları adım adım doğrulayabiliyor; ancak nihai kararı ve en yaratıcı kanıt yöntemlerinin keşfini hâlâ insan zekasına bırakıyor.
Peki ya bu denge değişirse?
#YapayZeka #Matematik #AlphaProof #Lean #mathlib #DeepMind #YaşamOyunu
https://youtu.be/ZERkY5kExjI
https://monologblg.com/matematik-sadece-insana-mi-ait/

Yapay zeka, büyük veri setlerini analiz etmekten karmaşık simülasyonlar oluşturmaya kadar, hesaplamalı matematiğin ağır işlerini üstlenerek büyük bir güç haline geldi. Şu an itibarıyla, YZ genellikle kendisine verilen ispatları adım adım doğrulayabiliyor; ancak nihai kararı ve en yaratıcı kanıt yöntemlerinin keşfini hâlâ insan zekasına bırakıyor.
Peki ya bu denge değişirse?
#YapayZeka #Matematik #AlphaProof #Lean #mathlib #DeepMind #YaşamOyunu
https://youtu.be/ZERkY5kExjI
https://monologblg.com/matematik-sadece-insana-mi-ait/

Yapay zeka, büyük veri setlerini analiz etmekten karmaşık simülasyonlar oluşturmaya kadar, hesaplamalı matematiğin ağır işlerini üstlenerek büyük bir güç haline geldi. Şu an itibarıyla, YZ genellikle kendisine verilen ispatları adım adım doğrulayabiliyor; ancak nihai kararı ve en yaratıcı kanıt yöntemlerinin keşfini hâlâ insan zekasına bırakıyor.
Peki ya bu denge değişirse?
#YapayZeka #Matematik #AlphaProof #Lean #mathlib #DeepMind #YaşamOyunu
https://youtu.be/ZERkY5kExjI
https://monologblg.com/matematik-sadece-insana-mi-ait/

Yapay zeka, büyük veri setlerini analiz etmekten karmaşık simülasyonlar oluşturmaya kadar, hesaplamalı matematiğin ağır işlerini üstlenerek büyük bir güç haline geldi. Şu an itibarıyla, YZ genellikle kendisine verilen ispatları adım adım doğrulayabiliyor; ancak nihai kararı ve en yaratıcı kanıt yöntemlerinin keşfini hâlâ insan zekasına bırakıyor.
Peki ya bu denge değişirse?
#YapayZeka #Matematik #AlphaProof #Lean #mathlib #DeepMind #YaşamOyunu
https://youtu.be/ZERkY5kExjI
https://monologblg.com/matematik-sadece-insana-mi-ait/

Readings shared June 10, 2025. https://jaalonso.github.io/vestigium/posts/2025/06/11-readings_shared_06-10-25 #AI #AIforMath #ATP #AlphaProof #Emacs #ITP #IsabelleHOL #LLMs #LeanProver #Math #Prover9

AlphaProof - Aprendizaje por refuerzo aplicado a la demostración matemática. https://jaalonso.github.io/vestigium/posts/2025/06/10-alphaproof-aprendizaje-por-refuerzo-aplicado-a-la-demostracion-matematica/ #AI #Math #AIforMath #ITP #LeanProver #AlphaProof

AlphaProof: When RL meets formal maths. ~ Thomas Hubert. https://www.youtube.com/live/xZIqn4V6O0A #AI #Math #AIforMath #ITP #LeanProver #AlphaProof

Readings shared April 3, 2025. https://jaalonso.github.io/vestigium/posts/2025/04/03-readings_shared_04-03-25 #AI #AlphaProof #CategoryTheory #FunctionalProgramming #Haskell #ITP #IsabelleHOL #LLMs #LeanProver #Math

AlphaProof: when reinforcement learning meets formal mathematics. ~ Thomas Hubert. https://youtu.be/TFBzP78Jp6A #AlphaProof #AI #Math

Readings shared November 29, 2024. https://jaalonso.github.io/vestigium/posts/2024/11/29-readings_shared_11-29-24 #ITP #Lean4 #IsabelleHOL #Logic #Math #Calculemus #MachineLearning #LLMs #AIMO #AlphaProof #AlphaGeometry #Haskell #FunctionalProgramming #Programming #FormalVerification #Dafny

Machine learning and mathematics (The twin catalysts poised to accelerate twenty-first century mathematics: formalisation and machine learning). ~ Harald Carlens. https://joltml.com/ml-mathematics/ #Math #ITP #Lean4 #MachineLearning #LLMs #AIMO #AlphaProof #AlphaGeometry

#ChatGPT geeft soms een perfecte oplossing als je hem een wiskundevraagstuk voorlegt, inclusief heldere uitleg bij tussenstappen. Maar vaak ook bluft hij maar wat. Het probleem met ChatGPT: hij weet zélf niet wanneer hij bluft.
Drie nieuwe AI-wiskundetools doen het beter: #AlphaProof, #AlphaGeometry en #o1. Mijn verhaal erover voor NRC
https://www.nrc.nl/nieuws/2024/10/04/ai-ontwikkelt-een-wiskundeknobbel-a4867967

Readings shared September 11, 2024. https://jaalonso.github.io/vestigium/posts/2024/09/11-readings_shared_09-11-24 #ITP #IsabelleHOL #LeanProver #Coq #NaprocheZF #Natty #AlphaGeometry #AlphaProof #FunctionalProgramming #OCaml #Logic #Math #IMO #AI #MachineLearning #LLMs

Mathematical Olympiad (To the geometry and beyond...). ~ Mirek Olšák. https://aitp-conference.org/2024/slides/MO.pdf #Math #ITP #AI #IMO #AlphaGeometry #AlphaProof #LeanProver

Readings shared August 20, 2024. https://jaalonso.github.io/vestigium/posts/2024/08/21-readings_shared_08-20-24 #ITP #Lean4 #Coq #ACL2 #AlphaProof #Math #Haskell #FunctionalProgramming #Logic #Math #CategoryTheory #TypeTheory

Improving on AlphaProof: IMO 2024 problem 2 in Lean 4. ~ David Renshaw. https://youtu.be/5IARsdn78xE #ITP #Lean4 #Math #AlphaProof

Deux #IA de #Google , baptisées #Alphaproof et #AlphaGeometry 2 :ces systèmes d' #IA spécialisés ont résolu quatre des six problèmes présentés cette année dans le cadre de l' Olympiade Internationale de Mathématiques
#Google revendique une percée dans ce domaine

https://www.generation-nt.com/actualites/ia-google-deepmind-mathematiques-raisonnement-2049320

AI reached the level of 12-year-old Terence Tao! Both won a Silver medal with 28 points.
#IMO #AI #Terence #Tao #TerenceTao #Math #Maths #Olympiad #Mathematics #SilverMedal #DeepMind #AlphaProof #AlphaGeometry

AI reached the level of 12-year-old Terence Tao! Both won a Silver medal with 28 points.
#IMO #AI #Terence #Tao #TerenceTao #Math #Maths #Olympiad #Mathematics #SilverMedal #DeepMind #AlphaProof #AlphaGeometry

Lecturas compartidas el 26 de julio de 2024. https://jaalonso.github.io/vestigium/posts/2024/07/27-lecturas_compartidas_el_26-jul-24 #ITP #Lean4 #IsabelleHOL #Coq #Math #IMO #AI #AlphaProof #FunctionalProgramming #Haskell

🧠 #DeepMind ha sviluppato due nuovi modelli di intelligenza artificiale, #AlphaProof e #AlphaGeometry 2, che sono riusciti a risolvere quattro dei sei problemi della International Mathematical Olympiad (IMO), raggiungendo il livello di una medaglia d'argento. 
💡 Questi modelli hanno dimostrato un avanzato ragionamento matematico, con AlphaProof specializzato in algebra e teoria dei numeri, mentre AlphaGeometry 2 ha eccelso in problemi geometrici. 

#AI #GenAI #GenerativeAI

Google DeepMind IMO 2024 Solutions. https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/imo-2024-solutions/index.html #AI #Math #IMO #Lean4 #AlphaProof #AlphaGeometry2

Google claims math breakthrough with proof-solving AI models - Enlarge / An illustration provided by Google. (credit: Google)

... - https://arstechnica.com/?p=2039221 #machinelearning #alphageometry2 #googledeepmind #alphageometry #alphaproof #biz⁢ #google #ai