The featured AI tool specializes in discerning whether a text is AI-generated, offering a probability output based on models like ChatGPT or Bard. Targeted at educators, SEO practitioners, and content reviewers, it aims to assist in distinguishing machine-generated content from human-created content.
This tool recommends inputting a minimum of 50 words for higher accuracy, providing a score showcasing the likelihood of AI generation for the entire text and highlighting sentences potentially created by AI due to low measures of perplexity. It underscores the necessity of combining its analysis with individual judgment due to potential false positives or negatives.
The AI detector functions on a machine learning system similar to those generating AI content, computing the probability of each word being AI-generated. It visualizes the results for the entire text and individual sentences, aiming to aid users in differentiating between human and AI-generated content.
It also offers an AI Content Detector extension for ChatGPT, enabling users to analyze text from any webpage, edit it, and recheck flagged sections for AI generation. However, it acknowledges the existence of false positives, especially in shorter, more generalized texts, pledging continual improvements for accuracy.
The tool's accuracy, highlighted through internal benchmarks, showcases a detection rate of over 97% for AI-generated texts while maintaining false positives below 3%. It notes that these benchmarks might differ from specific user texts.
Addressing text limitations, the free version processes texts up to 2000 characters, while pro and enterprise versions allow up to 8000 characters. Longer texts can be processed by breaking them into sections or utilizing the API.
Regarding compatibility, the AI check functions well with outputs from various language models, despite differences in architectures and datasets. It emphasizes the importance of detector updates to match evolving language models for optimal performance.
Share