Imagine teaching an AI to understand Finnish, a language renowned for its complex word structures. Researchers recently put cutting-edge Large Language Models (LLMs) to the test, challenging them to analyze Finnish words generated by a specialized tool. These words, far more intricate than those typically encountered, were designed to assess whether LLMs truly grasp the underlying rules of Finnish morphology or merely rely on statistical associations from their training data. The results were surprising. Even GPT-4-turbo, considered one of the most advanced LLMs, stumbled, exhibiting an incomplete understanding of Finnish grammar. While GPT-4 demonstrated some proficiency, it fell short of the accuracy achieved by smaller, specialized AI models trained specifically on Finnish morphology. Other powerful LLMs, like GPT-3.5-turbo, Llama 2-70B, and Poro-34B, struggled even more, revealing their limitations in handling the intricacies of Finnish word formation. This study underscores a critical challenge in AI research: while LLMs excel at generating human-like text, their comprehension of grammatical structures, especially in morphologically rich languages like Finnish, remains imperfect. The research points towards a potential reason for this deficiency: the way LLMs break down words into smaller units. This process, called tokenization, can sometimes obscure the morphological relationships between word parts, hindering the AI's ability to analyze complex forms. The study's findings have significant implications for the future development of LLMs. They highlight the need for improved methods that enable AIs to truly understand the grammatical structure of languages, moving beyond superficial pattern recognition to a deeper grasp of linguistic rules. This will be essential for developing truly multilingual AI systems capable of handling the diversity and complexity of human language.