This thesis explores a widespread issue with LLMs, which is Word Sense Disambiguation (WSD), focusing on ChatGPT. The study mainly focused on two sources of ambiguity: polysemy and homonymy, and how ChatGPT would perform in WSD in regards to these two sources in comparison with human reasoning. Using corpus-based examples and a controlled questionnaire, the study uncovers places where ChatGPT's structured approach to context can coincide with human judgement points, and also areas where human intuition handles finer semantic distinctions better than artificial intelligence. Besides illustrating ChatGPT’s increasing sophistication in language tasks, the findings also reveal occasional misclassifications made by it, thus serving to point out the complexity of contexts that are subtly nuanced. Thus, this thesis offers a rich portrayal of both the prospects and limitations of LLMs in WSD, calling again for model refinement in the future and collaborative human-AI approaches.