Pangram, a New York-based AI detection startup, has raised $9 million to address the growing issue of AI-generated content flooding the internet. The funding round, led by Menlo Ventures, includes participation from Haystack, ScOp, Script Capital, and Cadenza. Alongside the funding, Pangram launched its fourth-generation text detection model, Pangram 4, and an AI image detection model, Pangram Image. The text detection model boasts over 99% accuracy in identifying AI-assisted writing and mixed human-AI content, and it can more easily detect AI humanizer programs. The image detection model is currently available via research preview and will be released more widely in the coming weeks.
Pangram’s AI detection system is a large machine learning model trained on tens of millions of human documents. The startup created a 'synthetic mirror' for each document, replicating the topic, length, and tone of voice, but written by a frontier LLM. 'Our model is learning the stylistic differences and the choices that AI makes consistently and is able to use that to learn what makes something AI-generated with high confidence,' said Max Spero, one of the founders. The AI detector does not rely on copy-paste metadata or hidden watermarks, according to Spero.
Pangram’s AI detection system is designed not only to identify whether content was entirely written by AI but also to distinguish between levels of AI assistance. Spero believes AI assistance can be acceptable, provided the writer discloses its use. The startup’s technology is being used by Substack, Quora, schools, and universities, among others, according to Spero.
Source: techcrunch
