John Kirchenbauer of University of Maryland will be walking us through his paper A Watermark for Large Language Models this Wednesday.
The proposed watermark can be embedded with negligible impact on text quality, and can be detected using an efficient open-source algorithm without access to the language model API or parameters. It works by selecting a randomized set of "green" tokens before a word is generated, and then softly promoting use of green tokens during sampling.
Great office hours opportunity to dive into a trending paper with its one of its primary authors!
RSVP HERE: https://lu.ma/amo3or1a