Pangram released a system for detecting AI-generated text and images

Company Pangram introduced the fourth version of its tool for identifying content created by neural networks. The service supports analysis of both text fragments and graphic images, providing users with an assessment of the likelihood of machine generation.

Developers claim the accuracy of the algorithm's work at the level of 99.98%. To confirm the effectiveness of the system, the company involved experts from University of Maryland, who conducted an independent verification of the model's results. The verification methodology takes into account specific patterns characteristic of modern generative architectures.

Users have access to a free usage model, including 3 daily scans. The tool is focused on checking the authenticity of content in the context of the growing volume of synthetic data on the network. Details of the discussion of technical aspects of implementation are available in the branch of discussion on Hacker News.

The launch of Pangram 4 reflects the current trend towards creating specialized systems for controlling content quality. The implementation of such solutions allows for the automation of the moderation and verification process, reducing the load on manual verification systems in educational and media projects.