Paper Image

Detecting and evaluating watermarks in AI text generation

Published on:

28 March 2024

Primary Category:

Computation and Language

Paper Authors:

Piotr Molenda,

Adian Liusie,

Mark J. F. Gales


Key Details

Proposes comparative assessment to quantify watermark-based quality loss

Framework visualizes detection/quality tradeoff for watermark settings

Demonstrates analysis for multiple models and tasks

Shows high correlation to other evaluation methods

Discusses potential for transferring optimal settings

AI generated summary

Detecting and evaluating watermarks in AI text generation

This paper proposes a framework to analyze the tradeoff between watermark detection performance and quality degradation when watermarking large language models. Comparative assessment is used to quantify quality loss for different watermark settings across models and tasks.

Answers from this paper


No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up