The rise of large language models (LLMs) has revolutionized how we interact with technology, enabling impressive feats of language understanding and generation, especially in English. But what about languages with less digital presence? A new research paper explores this question by examining how well LLMs handle Cantonese, a language spoken by over 85 million people worldwide. The study reveals a significant gap in LLM development for Cantonese, particularly compared to languages from similarly developed regions. This disparity is concerning given the economic importance of Cantonese-speaking areas like the Guangdong-Hong Kong-Macau Greater Bay Area. Researchers constructed four new benchmarks to evaluate LLM performance in Cantonese across various tasks, including factual generation, mathematical logic, complex reasoning, and general knowledge. The results show that while some LLMs demonstrate promising abilities, overall performance in Cantonese lags behind English and Mandarin. One key challenge is the unique nature of Cantonese, including colloquialisms and code-switching between languages. For example, translating directly from English to Cantonese proves more difficult than translating from Mandarin due to closer linguistic similarities. However, the research also highlights opportunities. Data augmentation techniques—creating more training data by translating existing text from other languages—offer a path to enhance LLM performance in Cantonese. Certain LLMs were identified as more effective for this translation process. Another key area is improving how LLMs handle code-switching, the common practice of blending Cantonese with English and other languages in everyday conversation. This research underscores the need for greater focus on underrepresented languages in the AI field. Addressing these challenges will not only benefit Cantonese speakers but will also drive advancements in natural language processing for diverse languages around the globe. Future development of multilingual and code-switching capabilities in LLMs has far-reaching implications for communication, education, and cross-cultural understanding in our increasingly interconnected world.