Technical Report on the Pangram AI-Generated Text Classifier
This work addresses the need for reliable AI text detection across various domains, offering a significant improvement over current tools, though it is incremental in advancing detection accuracy.
The researchers tackled the problem of distinguishing AI-generated text from human-written text by developing Pangram Text, a transformer-based classifier that achieved over 38 times lower error rates compared to existing methods on a comprehensive benchmark across 10 text domains and 8 language models.
We present Pangram Text, a transformer-based neural network trained to distinguish text written by large language models from text written by humans. Pangram Text outperforms zero-shot methods such as DetectGPT as well as leading commercial AI detection tools with over 38 times lower error rates on a comprehensive benchmark comprised of 10 text domains (student writing, creative writing, scientific writing, books, encyclopedias, news, email, scientific papers, short-form Q&A) and 8 open- and closed-source large language models. We propose a training algorithm, hard negative mining with synthetic mirrors, that enables our classifier to achieve orders of magnitude lower false positive rates on high-data domains such as reviews. Finally, we show that Pangram Text is not biased against nonnative English speakers and generalizes to domains and models unseen during training.