Loading...
Loading...
Browse all stories on DeepNewz
VisitOpenAI AI Model Scores Over 90% on MATH Dataset, Available by 2026
Jul 15, 2024, 04:57 PM
OpenAI has reportedly tested an AI model internally that scored over 90% on a MATH dataset, which is a benchmark of championship math problems. This development suggests significant progress in AI capabilities, potentially linked to the 'Strawberry' project. Other AI models, such as Deepseek-Math (7B), Gemini 1.5 Pro (May), and GPT-4o, scored 51.7%, 67.7%, and 76.6% respectively. The high score achieved by OpenAI's model indicates a breakthrough in AI performance, though some experts caution about overfitting on such datasets. The model is predicted to be available by 2026 and may involve the Q* algorithm.
View original story
Markets
Yes • 50%
No • 50%
Official announcement from OpenAI or a reputable news source confirming the release and performance of the model
No • 50%
Yes • 50%
Official announcements from educational platforms or OpenAI
No • 50%
Yes • 50%
Official records or announcements from major international math competitions
Recurrent Neural Network-based • 25%
Other • 25%
Q* algorithm • 25%
Transformer-based • 25%
Official technical papers or announcements from OpenAI
Gemini 1.5 Pro (May) • 25%
OpenAI's model • 25%
GPT-4o • 25%
Deepseek-Math (7B) • 25%
Official publications or announcements from AI research organizations
GPT-4o • 25%
OpenAI's model • 25%
Deepseek-Math (7B) • 25%
Gemini 1.5 Pro (May) • 25%
Official publications or announcements from AI research organizations