GLTR

AI content detectionAI content detection

Description

"GLTR is an AI tool developed by MIT-IBM Watson AI Lab and HarvardNLP for detecting automatically generated text. It analyzes the likelihood of text being generated by a language model, visually highlighting words based on their probability. Detect fake reviews, comments, or news articles easily with GLTR. Access the live demo and source code on Github."

About GLTR

Welcome to GLTR, an innovative tool developed by the MIT-IBM Watson AI lab and HarvardNLP. Our goal is to provide you with the ability to detect automatically generated text using forensic analysis. With GLTR, you can easily determine if a text has been artificially generated through a process of thorough evaluation.

Using advanced analysis techniques, GLTR examines the likelihood of a language model generating a specific text. We specifically focus on the GPT-2 117M language model from OpenAI, leveraging its capabilities to analyze and rank each word based on its probability of being produced by the model.

When analyzing a text, GLTR color codes the words to enable quick identification. The most likely words are highlighted in green, followed by yellow and red. The remaining words are depicted in purple. This direct visual indication allows you to swiftly identify computer-generated text within any given document.

Furthermore, GLTR provides three insightful histograms that aggregate information about the entire text. The first histogram showcases the number of words in each category, offering an overview of the distribution. The second histogram highlights the ratio between the probabilities of the top predicted word and the following word, providing crucial insights. Lastly, the third histogram presents the distribution of entropy among the predictions.

By carefully analyzing these histograms, GLTR provides additional evidence to help you determine if a text has been artificially generated. Its functionalities make it an invaluable tool for identifying fake reviews, comments, or news articles created by large language models. These models have the potential to produce texts that are virtually indistinguishable from human-written content, particularly for non-expert readers.

Are you curious to give GLTR a try? Access our live demo to experience its capabilities firsthand. For researchers interested in a closer look, the source code is available on Github. Additionally, you can explore the ACL 2019 demo track paper, which was nominated for the best demo award.

Tags

AI content detection
Share tool
    Get product updates
    Be the first to try new Tellit features