We live in a world where machines are writing like humans. From crafting tweets to generating news articles, large language models (LLMs) are blurring the lines between human and machine-authored content. But how can we tell if a text was written by a human or an AI? This question has become increasingly important as AI-generated text proliferates, potentially leading to misinformation and misuse. New research explored this challenge by examining how well traditional machine learning methods can detect machine-generated content across various text lengths. In this study, the researchers analyzed four datasets containing everything from short tweets to long-form web text. They found that detecting text from larger LLMs (like GPT-2's XL variant) is harder than detecting text from smaller models. Traditional methods achieved around 96% accuracy with smaller models but only around 74% with the largest one. The study also examined the characteristics of human and machine-generated text across multiple dimensions. Machine-generated text tended to be more readable but less emotionally expressive. It sometimes also exhibited a stronger bias, raising ethical concerns. Interestingly, the study found that machine-generated text could closely mimic human moral judgments, suggesting a need for further research into the ethical implications of LLMs. Furthermore, the study confirmed the difficulty of detecting rephrased texts. Models struggled more when machine-generated text closely mimicked the style and tone of human-written content. The research highlights that while machines are getting better at writing, traditional machine learning methods can still effectively detect AI-generated text in many cases. The research also highlights ongoing challenges in detecting content from larger models and rephrased text, indicating areas where future research is needed. As AI writing becomes more sophisticated, the importance of detection methods will only continue to grow.