OpenAI Unveils Multilingual AI Benchmark Dataset on Hugging Face

The MMMLU dataset evaluates language models across 14 languages, promoting inclusivity and fairness in AI research.

OpenAI's MMMLU dataset tests AI models on tasks requiring general knowledge, reasoning, problem-solving, and comprehension in multiple languages.
The dataset includes questions from various subjects, ranging from high-school level to advanced professional knowledge, ensuring a comprehensive evaluation.
Languages like Swahili, Yoruba, and Bengali are part of the dataset, addressing the underrepresentation of low-resource languages in AI research.
MMMLU was created using professional human translators to ensure higher accuracy compared to datasets relying on machine translation.
By releasing the dataset on Hugging Face, OpenAI aims to enhance open access to advanced AI tools and support global AI research.