Top Tweets for #AlphaGeometry

about 1 year ago

Taking a short break ... but will soon be back to our deep thinking mode! Thanks to @demishassabis for featuring some of our works from IMO 2024 (#AlphaGeometry, #AlphaProof) to #DeepThink + other amazing works from colleagues at @GoogleDeepMind! https://t.co/4O7jM4HLLU

lmthang's tweet photo. Taking a short break ... but will soon be back to our deep thinking mode! Thanks to @demishassabis for featuring some of our works from IMO 2024 (#AlphaGeometry, #AlphaProof) to #DeepThink + other amazing works from colleagues at @GoogleDeepMind! https://t.co/4O7jM4HLLU https://t.co/A7yHgsky1g

about 1 year ago

And our team has been deep in deep thinking ...

15

1K

15

73

105K

1

33

1

2

3K

over 1 year ago

Gold-medalist performance in solving olympiad geometry with AlphaGeometry2. ~ Yuri Chervonyi et als. https://t.co/uvIJpM4LXi #AI #AlphaGeometry #Math

0

15

2

5

711

AIopenmind @AIopenmind

over 1 year ago

🏅 AlphaGeometry2 di Google DeepMind supera la media dei vincitori IMO : l'IA risolve l'84% dei problemi di geometria olimpionica grazie a un motore simbolico avanzato ea Gemini. #AI! 🔢🤖 #AlphaGeometry #DeepMind #MathOlympiad https://t.co/9sWNMASCIr

0

24

over 1 year ago

Machine learning and mathematics (The twin catalysts poised to accelerate twenty-first century mathematics: formalisation and machine learning). ~ Harald Carlens. https://t.co/X4Pm88hXdj #Math #ITP #Lean4 #MachineLearning #LLMs #AIMO #AlphaProof #AlphaGeometry

0

15

4

2K

almost 2 years ago

Readings shared September 11, 2024. https://t.co/yGKdi0D1Ji #ITP #IsabelleHOL #LeanProver #Coq #NaprocheZF #Natty #AlphaGeometry #AlphaProof #FunctionalProgramming #OCaml #Logic #Math #IMO #AI #MachineLearning #LLMs

0

1

0

261

Prof. Dr. Marco Barenkamp @m_barenkamp

almost 2 years ago

Mathematical Olympiad (To the geometry and beyond...). ~ Mirek Olšák. https://t.co/uDikmDHN5y #Math #ITP #AI #IMO #AlphaGeometry #AlphaProof #LeanProver

0

7

1

2

410

Eden AI @EdenAiza

almost 2 years ago

The potential of #AI solving challenging mathematical problems realised in new Google DeepMind's AI systems, #AlphaProof & #AlphaGeometry 2 tackled 4 out of 6 problems at the 2024 International Mathematical Olympiad! Read more: https://t.co/YfcbUVjT75 #edenai #GPT #GPT4 #latest

EdenAiza's tweet photo. The potential of #AI solving challenging mathematical problems realised in new Google DeepMind's AI systems, #AlphaProof & #AlphaGeometry 2 tackled 4 out of 6 problems at the 2024 International Mathematical Olympiad!
Read more: https://t.co/YfcbUVjT75

#edenai #GPT #GPT4 #latest https://t.co/fUv27JnuY9

0

4

1

0

138

DongNan @liuxiaodong9630

almost 2 years ago

Domain name is for sale https://t.co/pgzVeXGxCd #AlphaGeometry #ai #math #GoogleProject2025 #agi #domainsforsale #domainname

liuxiaodong9630's tweet photo. Domain name is for sale
https://t.co/pgzVeXGxCd

#AlphaGeometry #ai #math
#GoogleProject2025 #agi
#domainsforsale #domainname https://t.co/DG8jUEj2JV

0

2

0

59

almost 2 years ago

#Google-#KI gewinnt Silber bei der #Mathematik-Olympiade: Vier von sechs Aufgaben zu beantworten reichte dabei aus. Bei der Lösung von einer der sechs Aufgaben haben die KIs #AlphaProof und #AlphaGeometry 2 allerdings drei Tage benötigt. Via @heiseonline: https://t.co/tQprJ3cZhH

0

3

2

1

188

Express Technology @ExpressTechie

almost 2 years ago

The current class of AI models, which work by statistically predicting the next word, have struggled with abstract math, which requires greater reasoning capabilities resembling human intelligence. #Google #AI #AlphaGeometry https://t.co/Bw6eesssVN

0

3

0

1K

almost 2 years ago

We created #AlphaGeometry 2 with lots of innovations: (a) one order of magnitude more synthetic data for the language model (Gemini-based, trained from scratch); (b) a symbolic engine that are two orders of magnitude faster, and (c) a novel knowledge-sharing mechanism to tackle more complex problems. We can now solve 83% of IMO geometry problems compared to our previous rate of 53%. For the geometry problem in #imo2024, no problem, we solved it in 19s! 🚀 Thanks to @thtrieu_ and the amazing AlphaGeometry team! IMO problem #4 asked to prove that the two angles (in blue) sum to 180°. Point E (in red) is what AlphaGeometry 2 proposed, which looked strange at first to our experts, but then turned out to give purpose to the midpoint L of AB, eventually leading to prove the conclusion. To me, the point E feels like “move 37” by AlphaGo as it wasn't intuitive to experts at first, but connects many other pieces together.

lmthang's tweet photo. We created #AlphaGeometry 2 with lots of innovations: (a) one order of magnitude more synthetic data for the language model (Gemini-based, trained from scratch); (b) a symbolic engine that are two orders of magnitude faster, and (c) a novel knowledge-sharing mechanism to tackle more complex problems. We can now solve 83% of IMO geometry problems compared to our previous rate of 53%. For the geometry problem in #imo2024, no problem, we solved it in 19s! 🚀 Thanks to @thtrieu_ and the amazing AlphaGeometry team!

IMO problem #4 asked to prove that the two angles (in blue) sum to 180°. Point E (in red) is what AlphaGeometry 2 proposed, which looked strange at first to our experts, but then turned out to give purpose to the midpoint L of AB, eventually leading to prove the conclusion. To me, the point E feels like “move 37” by AlphaGo as it wasn't intuitive to experts at first, but connects many other pieces together.

1

15

1

4

3K

almost 2 years ago

Neuro-symbolic paradigm strikes again! If in #AlphaGeometry, the creative language model (System 1) suggests insights for the reliable symbolic engine (System 2) to complete a proof, we see that pattern again in #AlphaProof. The language model suggests key proof steps in a special language (called Lean); the Lean engine then takes care of the rest. Big congrats to the AlphaProof team for having successfully applied AlphaZero approach to formal Math. The trend will continue: language model + search will be a crucial recipe for advanced reasoning, something that I believe is the key prerequisite for AGI!

lmthang's tweet photo. Neuro-symbolic paradigm strikes again! If in #AlphaGeometry, the creative language model (System 1) suggests insights for the reliable symbolic engine (System 2) to complete a proof, we see that pattern again in #AlphaProof. The language model suggests key proof steps in a special language (called Lean); the Lean engine then takes care of the rest. Big congrats to the AlphaProof team for having successfully applied AlphaZero approach to formal Math. The trend will continue: language model + search will be a crucial recipe for advanced reasoning, something that I believe is the key prerequisite for AGI!

2

38

3

10

4K

almost 2 years ago

Super thrilled to share that our AI has has now reached silver medalist level in Math at #imo2024 (1 point away from 🥇)! Since Jan, we now not only have a much stronger version of #AlphaGeometry, but also an entirely new system called #AlphaProof, capable of solving many more Olympiad problems. This is a large-scale project that I was fortunate to co-lead at @GoogleDeepMind! See our blog & NYT articel below! Blog: https://t.co/CgsTtLOEN6 NYT: https://t.co/h0TkPNwfeL

14

237

29

45

475K

about 2 years ago

Glad to see us keep pushing the frontier on maths! Also a bit of a teaser to IMO-Bench (after #alphageometry), which my team (with @quocleix's support) built & the best model only got 25% :) Check out the cool answer of the Gemini Math on an APMO problem in @OriolVinyalsML's thread below, which is still an easy problem in the #imobench ;)

Oriol Vinyals

@OriolVinyalsML

about 2 years ago

Today we have published our updated Gemini 1.5 Model Technical Report. As @JeffDean highlights, we have made significant progress in Gemini 1.5 Pro across all key benchmarks; TL;DR: 1.5 Pro > 1.0 Ultra, 1.5 Flash (our fastest model) ~= 1.0 Ultra. As a math undergrad, our drastic results in mathematics are particularly exciting to me! In section 7 of the tech report, we present new results on a math-specialised variant of Gemini 1.5 Pro which performs strongly on competition-level math problems, including a breakthrough performance of 91.1% on Hendryck’s MATH benchmark without tool-use (examples below 🧵). Gemini 1.5 is widely available, try it out for free here https://t.co/GJXW8lduNk & read the full tech report here: https://t.co/Pltp92WcNo

OriolVinyalsML's tweet photo. Today we have published our updated Gemini 1.5 Model Technical Report. As @JeffDean highlights, we have made significant progress in Gemini 1.5 Pro across all key benchmarks; TL;DR: 1.5 Pro > 1.0 Ultra, 1.5 Flash (our fastest model) ~= 1.0 Ultra.

As a math undergrad, our drastic results in mathematics are particularly exciting to me!

In section 7 of the tech report, we present new results on a math-specialised variant of Gemini 1.5 Pro which performs strongly on competition-level math problems, including a breakthrough performance of 91.1% on Hendryck’s MATH benchmark without tool-use (examples below 🧵).

Gemini 1.5 is widely available, try it out for free here https://t.co/GJXW8lduNk & read the full tech report here: https://t.co/Pltp92WcNo

40

985

188

315

714K

1

38

3

5

7K