GLTR

GLTR (Giant Language model Test Room) is a statistical tool for detecting machine-generated text. The system processes user-submitted text and analyzes it against a language model's predictions. For each word, it determines its probability of appearing given the preceding context. The output is a visual overlay on the original text, color-coding each word based on its rank in the model's probability distribution. A histogram also displays the distribution of these word probabilities, providing a statistical signature to help distinguish between human and machine authorship. The tool is intended for professionals who must assess text authenticity, including content strategists, SEO specialists, and academic integrity officers. A typical use case involves an SEO specialist pasting competitor or user-generated content into the interface to evaluate if it was machine-generated, which can inform content strategy or moderation. Another application is for editors and educators to analyze submitted articles or assignments, using the statistical and visual feedback to identify passages that may warrant further scrutiny for AI-assisted writing.