Words are like chameleons, constantly shifting their colors—or meanings—over time. "Sick" goes from unwell to awesome, "cloud" transforms from fluffy to digital, all thanks to the subtle dance of semantic change. But how can we teach machines to understand these ever-evolving linguistic landscapes? Researchers are exploring the fascinating power of Large Language Models (LLMs) to not only detect these shifts but to characterize *how* words change meaning. This groundbreaking research combines the reasoning abilities of LLMs with the art of rhetoric, using figures of speech like zeugmas and similes to pinpoint the nuances of semantic change. Imagine an AI dissecting the word "cup"—does it mean a container, or a unit of measurement? By attempting to create a grammatically correct yet semantically awkward sentence like "He drank a cup of sugar," the model detects the difference. Similarly, LLMs can unravel the difference between literal and metaphorical meanings. When "battle" describes a war and then a boardroom debate, the model uses rhetorical analysis to categorize this shift as a metaphor. This approach reveals the impressive potential of LLMs to not just parrot words, but to deeply understand their context and the subtle shifts in meaning that shape our communication. The next step is to explore the limitations of these models and to make their reasoning even more transparent, paving the way for AI systems that can truly grasp the complexities of language evolution.