LLMs such as Chat Generative Pre-Trained Transformer (ChatGPT), Pathways Language Models (PaLM), and Bard Artificial Intelligence (Bard AI) can generate human-like text. On top of that, interestingly, these generated texts can correspond to ‘any’ domain of human life such as finance, medicine, and health. Due to the training of these LLMs on a large amount of text corpus, the generated text is hard to be detected merely by reading. Therefore, it creates a havoc of privacy and copyright issues and hence effective detection of the generated texts is critical. In this research, we take a strong step towards effectively detecting generated text that belongs to several domains humans deal with in our day-to-day lives. Interestingly, existing detection methods utilize either LLM models or deep neural networks to detect generated text; one catch here is that since these models also utilize similar backbones, therefore, existing defenses are found non-generalized. We further categorize the difficult examples into multiple categories based on the evaluation settings such as unseen domain, unseen dataset, and modified text, and show the detection performance of different categories. In this research, we utilize the potential of traditional machine learning classifiers to differentiate human text from generated text in a resource-efficient manner. Our extensive investigation reveals surprising yet simple relationships between generated examples from different domains. It is demonstrated that paraphrasing can degrade the performance of LLM-based AI-generated text detection algorithms drastically as compared to the traditional classifiers. Further, the fusion of multiple text encoders and machine learning classifiers can boost the performance of a single model.