Specifically, we first define a binary encoding function to compute a random binary encoding corresponding to a word. To allow third-parties to autonomously inject watermarks into generated text, we develop a watermarking framework for black-box language model usage scenarios. For instance, third-parties that develop API-based vertical applications cannot watermark text themselves because API providers only supply generated text and withhold probability distributions to shield their commercial interests. However, this method is not applicable in many real-world scenarios where only black-box language models are available. A detection algorithm aware of the list can identify the watermarked text.
The method involves randomly dividing the model vocabulary to obtain a special list and adjusting the probability distribution to promote the selection of words in the list. To achieve reliable detection, a watermark-based method was proposed for white-box LLMs, allowing them to embed watermarks during text generation. However, passive detection methods are stuck in domain specificity and limited adversarial robustness. Thus, detecting generated text is crucial.
BLACK BOXES MICROSOFT WORD HEADING PDF
Download a PDF of the paper titled Watermarking Text Generated by Black-Box Language Models, by Xi Yang and 7 other authors Download PDF Abstract:LLMs now exhibit human-like skills in various fields, leading to worries about misuse.
0 kommentar(er)
