A watermark for chatbots can spot text written by an AI
For example, since OpenAI’s ChatGPT chatbot was launched in November, students have started cheating using it to write essays for them. CNET news site used ChatGPT to write articles, just for release corrections Between accusations Plagiarism. Building watermarking into such systems before they are released can help with such problems.
In studies, these watermarks have been used to identify AI-generated text with near certainty. For example, researchers at the University of Maryland were able to detect text generated by Meta’s open source language model, OPT-6.7B, using a detection algorithm they built. The job is described in a paper that hasn’t been peer-reviewed yet, and code will be available Free around February 15th.
AI language models work by predicting and generating words one by one. After each word, the watermarking algorithm randomly divides the language model’s vocabulary into words from the “green list” and “red list”, and then prompts the model to select the words from the green list.
The more greenlisted words in a paragraph, the more likely it is that the text is machine generated. Text written by one person tends to contain more random mixed words. For example, for the word “beautiful”, the watermarking algorithm can classify the word “flower” as green and “orchid” as red. Tom Goldstein, an assistant professor at the University of Maryland who was involved in the study, explains: An AI model with a watermarking algorithm would be more likely to use the word “flower” rather than “orchid”.