Large language models (LLMs) have made incredible strides in understanding and generating text, but they often stumble when it comes to cultural nuances. Their training data, frequently dominated by Western perspectives, can lead to inaccurate or culturally biased responses, especially in non-English languages. This raises a critical question: how can we build AI that truly understands the world's diverse cultures? Researchers are tackling this challenge by developing culturally-aligned datasets to evaluate and fine-tune LLMs. One promising approach is NativQA, a framework designed to create natural question-answering datasets in multiple languages, reflecting local knowledge and cultural contexts. The team behind NativQA has demonstrated its potential by creating MultiNativQA, a dataset with over 64,000 question-answer pairs in seven languages, ranging from widely spoken to extremely low-resource. The questions, gathered from native speakers in nine regions, cover a variety of everyday topics. This approach helps reveal the blind spots of existing LLMs when confronted with culture-specific queries. Early experiments using MultiNativQA show that fine-tuning an LLM with this dataset can significantly improve its performance in low-resource languages and dialects, showcasing the value of incorporating native cultural knowledge. The team is making both the framework and the dataset publicly available, fostering collaboration and further research in building truly multilingual and culturally-aware AI. The work on NativQA and MultiNativQA highlights a key challenge in AI development: moving beyond generic language understanding to a more nuanced and inclusive approach. While the current dataset focuses on a select group of languages and regions, it provides a valuable blueprint for future work. As LLMs become increasingly integrated into our lives, datasets like this will be crucial in ensuring they are not only intelligent, but also culturally aware and sensitive to the world's rich diversity.