BLEnD: Benchmark for LLMs on Everyday Knowledge in Diverse Cultures and Languages

BLEnD is a benchmark for evaluating large language models on culture-specific everyday knowledge across diverse regions and non-English languages. The dataset was submitted to the NeurIPS 2024 Datasets and Benchmarks Track by author nayeon212. It was last updated on May 2, 2025.

Use Cases

• Benchmarking LLM performance on culture-specific knowledge based on the described focus on everyday knowledge.
• Evaluating multilingual reasoning in non-English languages as suggested by the dataset's purpose.
• Analyzing geographic or cultural bias in LLMs based on the dataset's coverage of diverse regions.

Strengths

• Dataset is designed for a specific, documented research purpose: benchmarking LLMs on cultural knowledge.
• Focuses on a recognized gap in LLM evaluation: everyday knowledge in diverse cultures and languages.
• Maintained with updates, including a revision to multiple choice questions noted on 2025-05-02.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and overall data scale are unknown, which may limit suitability assessment.
• The full description is hosted externally, requiring a page visit for complete context.

Provenance

Source

nayeon212 on Hugging Face

Freshness

Last updated 2025-05-02 14:55:48

Geography

Diverse regions (inferred from description)