The rise of sophisticated AI writing tools has sparked a crucial question: can we reliably detect when a text is machine-generated? This challenge is even trickier when dealing with *partially* AI-written content, where human and machine text intertwine. A new research project tackled this complex problem by developing a model to pinpoint the exact word where machine generation begins in a piece of text. Researchers trained their model on a dataset of partially human-written academic essays and peer reviews, some completed by AI models like ChatGPT, LLaMA 2, and GPT-4. They focused on identifying the “text boundary”—the precise shift from human to machine writing. The model employed a combination of powerful language models like DeBERTa and a technique called Conditional Random Fields (CRF) to analyze text at the word level. Impressively, the DeBERTa-CRF model achieved near-perfect accuracy on both seen and unseen datasets, suggesting its ability to generalize across different writing styles and AI generators. It even outperformed commercial AI detection systems, especially with shorter texts. However, challenges remain. The model showed some weaknesses when text boundaries fell in the middle of a sentence and struggled with certain grammatical patterns where training data was sparse. Improving the model involves gathering more diverse training data (including different text types like social media posts) and exploring smarter ways to combine the strengths of different language models. This research has important implications for combating misinformation, ensuring academic integrity, and building trust in online content. As AI writing becomes more prevalent, accurately detecting machine-generated text, especially within partially human-written content, will be critical. Further research might explore multilingual applications, analyze the model's effectiveness against paraphrased machine-generated text, and investigate cases where multiple AI generators are used within the same text. This research is a vital step in the ongoing effort to navigate the increasingly complex landscape of human and machine-generated text.