Thomas

Thomas provides a specialized tool for detecting whether a piece of text has been generated by GPT models such as ChatGPT, GPT-4, or other large language models. Users can paste or upload text content into the detection interface, and the system analyzes multiple linguistic and statistical features to determine the likelihood of AI authorship. The detection engine examines patterns like sentence length variability, token repetition, perplexity scores, and burstiness — characteristics that often differ between human-written and machine-generated text. It also compares the input against known distributions of GPT outputs. The tool supports multiple languages including English, Chinese, Spanish, French, and German, making it useful for educators who want to verify student submissions, content moderators checking for automated spam, and researchers studying AI text generation. Results are presented as a probability score along with a highlighted breakdown of suspicious segments. The service also offers an API endpoint for batch processing, allowing integration into plagiarism detection systems, publishing platforms, and content management workflows. Privacy is handled by not storing uploaded texts permanently; each analysis session is ephemeral. The interface is minimal and focuses on one function: accurate detection without unnecessary features. It is used by universities to maintain academic integrity, by newsrooms to flag AI-written articles, and by freelance writers to self-check before submission. The detection model is periodically retrained on new GPT releases to keep up with evolving generation techniques. Users can also adjust sensitivity thresholds depending on their tolerance for false positives. Detailed documentation explains the methodology behind each metric.