Advancing Beyond Identification: Multi-bit Watermark for Large Language Models

arXiv:2308.00221v3 Announce Type: replace
Abstract: We show the viability of tackling misuses of large language models beyond the identification of machine-generated text. While existing zero-bit watermark methods focus on detection only, some malicious misuses demand tracing the adversary user for counteracting them. To address this, we propose Multi-bit Watermark via Position Allocation, embedding traceable multi-bit information during language model generation. Through allocating tokens onto different parts of the messages, we embed longer messages in high corruption settings without added latency. …

arxiv beyond cs.ai cs.cl cs.cr identification language language models large language large language models type watermark