Ghostbuster: Detecting Text Ghostwritten by Large Language Models

Ghostbuster is a state-of-the-art method for detecting AI-generated text. It works by finding the probability of generating each token in a document under weaker language models and combining these probabilities as input to a final classifier. Ghostbuster achieves high performance in detecting AI-generated text across different domains and models, and it doesn't require knowing the specific model used to generate the text. Future directions for Ghostbuster include providing explanations for model decisions and improving robustness to attacks.