Begin typing your search above and press return to search.
proflie-avatar
Login
exit_to_app
DEEP READ
Ukraine
access_time 16 Aug 2023 5:46 AM
Putin
access_time 2 Jan 2025 8:06 AM
What is Christmas?
access_time 26 Dec 2024 5:49 AM
Munambam Waqf issue decoded
access_time 16 Nov 2024 5:18 PM
exit_to_app
Homechevron_rightTechnologychevron_rightGoogle says its AI...

Google says its AI outperformed Math Olympiad gold medalists in Geometry

text_fields
bookmark_border
Google DeepMind
cancel

Google has announced a groundbreaking achievement in artificial intelligence, revealing that its latest AI system, AlphaGeometry2, has surpassed the problem-solving abilities of gold medalists at the International Mathematical Olympiad (IMO).

The AI system, developed by DeepMind, successfully solved 84% of complex geometry problems from past IMO competitions, outperforming the 81.8% success rate of average human gold medalists. Given the challenging nature of IMO problems, which require deep mathematical reasoning, this marks a significant advancement in AI’s ability to handle complex mathematical concepts.

Initially introduced in January last year, AlphaGeometry had performed at the level of IMO silver medalists. However, over the past year, Google has significantly enhanced the system’s capabilities.

To improve problem-solving accuracy, Google expanded AlphaGeometry's mathematical language, enabling it to tackle more intricate geometric problems that involve object movements, angles, ratios, and distances. "This enhancement increased AlphaGeometry’s ability to cover IMO geometry problems from 66% to 88%, significantly boosting its problem-solving range," Google stated.

Additionally, the company integrated its advanced Gemini AI tool to refine the system’s language modeling and search processes, further improving its efficiency.

One of the most notable upgrades allows AlphaGeometry2 to manipulate geometric objects dynamically. For instance, the AI can now move a point along a line to adjust a triangle’s height, or effectively solve linear equations—a capability that was previously limited.

Despite these impressive achievements, Google acknowledges that AlphaGeometry2 is not yet perfect.

"Our domain language still lacks the ability to handle problems involving a variable number of points, non-linear equations, and inequalities. Addressing these gaps will be crucial to achieving full automation in geometry problem-solving," DeepMind researchers explained.

Looking ahead, Google's long-term plan for AlphaGeometry2 is to achieve error-free, fully automated geometry problem-solving. The researchers also aim to speed up the AI’s inference process, making it even more reliable and efficient.

Show Full Article
TAGS:Google DeepMindAlphaGeometry2
Next Story