Shypd.aiThe largest open index of AI toolsCatching Unicorns with GLTR is an open-source tool that helps detect AI-generated text. It provides a visual footprint of language models on input text, highlighting words based on their likelihood of being predicted by a model like GPT-2.
Trending
About
Catching Unicorns with GLTR is a forensic tool designed to inspect the visual footprint of language models on input text, helping users determine whether a text was likely generated by an AI. Developed in collaboration by MIT-IBM Watson AI lab and HarvardNLP, GLTR uses the GPT-2 117M language model to analyze text. It overlays a colored mask on the text, highlighting words in green (top 10 most likely), yellow (top 100), red (top 1,000), and purple (rest) based on their predicted rank. This visual representation allows for a forensic analysis of how likely an automatic system has generated a text, as AI-generated content tends to stick to the most probable words. The tool also provides histograms to aggregate information on word categories, probability ratios, and prediction entropies, offering deeper insights into the text's origin. It's particularly useful for identifying texts that appear "too likely" to be human-written, as natural writing often includes more unpredictable words.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
