(Authored by Or Zamir, Tel Aviv University, [email protected])
Full paper: [https://arxiv.org/abs/2401.10360]
The main file is generate.py, the main method in which generates the plot and figures from the paper. This code was not written, designed, optimized or tested for any use or application besides research. Commercial use is not allowed.
Acknowledgments: The base code we used for generating an LLM response token-by-token is adapted from the following repository [https://github.com/jthickstun/watermark/] that was written by Rohith Kuditipudi, John Thickstun, Tatsunori Hashimoto, and Percy Liang.
Cite as:
@misc{zamir2024excuse,
title={Excuse me, sir? Your language model is leaking (information)},
author={Or Zamir},
year={2024},
eprint={2401.10360},
archivePrefix={arXiv},
primaryClass={cs.CR}
}