Large language models (LLMs) are everywhere, but how can we tell if a piece of text was written by a human or generated by a machine? Researchers have developed a clever new method called SPOT (Source Prediction from Originality Score Thresholding) that tackles this challenge. Instead of focusing on the content's truthfulness, SPOT looks at its *originality*. It works by using one LLM to analyze text and predict how likely another LLM would be to generate the same words. If the first LLM finds the text highly predictable, it's likely machine-generated. But if the text is unpredictable, it's more likely to be human-written. Tests show SPOT is surprisingly effective across different LLMs, sizes, and datasets. It even works with compressed models, making it practical for real-world use. However, SPOT isn't foolproof. It struggles with specialized text like code or math, where LLMs excel due to the narrower vocabulary and predictable patterns. Also, it hasn't been tested on text that's a mix of human and AI writing. Despite these limitations, SPOT offers a promising new tool for identifying AI-generated text, especially in everyday applications like email filtering. It raises interesting questions about how we define originality and what it means for the future of writing in an AI-driven world.