Openai recently released a new version of the GPT-4-04-09 chatbot, which has already demonstrated its improved abilities, especially in solving the mathematical tasks of the Olympiad level. The update has significantly surpassed its predecessors in this area, demonstrating impressive effectiveness.
Casted tests on the Kaggle platform as part of the competition ai Mathematical Olympiad PRIZE showed that the updated GPT -4 was able to successfully solve 5 out of 10 proposed tasks. The previous version coped with only 1 task, and to solve it, a request with hints was required. As a comparison, the competitor Claude 3 Opus was able to solve only 1 problem without hints.
The final results of the test
The progress of the Chat Bota was highly appreciated by Openai, calling its “essential”, especially emphasizing the improvement in solving mathematical problems.
Openai employee about the new version of the GPT-4
It is expected that the new version will be integrated into the ChatGPT web version with the subscription, and is also presented on the LMSYS Arena platform, where users will be able to directly evaluate the improvements.
It should be noted that the release of the new version of the GPT-4 coincided with the update of the rating of intellectual models, where the CMD-R+ model from Cohere with parameters of more than 100 billion surpassed the previous versions of GPT-4 from the spring of 2023. Nevertheless, Claude 3 OPUS still holds a leading position in the ranking. In the near future, publications metrics that will demonstrate the quality of reasoning and abilities in mathematics of the new version of GPT-4.