Copyright Traps for Large Language Models

arXiv:2402.09363v1 Announce Type: new
Abstract: Questions of fair use of copyright-protected content to train Large Language Models (LLMs) are being very actively debated. Document-level inference has been proposed as a new task: inferring from black-box access to the trained model whether a piece of content has been seen during training. SOTA methods however rely on naturally occurring memorization of (part of) the content. While very effective against models that memorize a lot, we hypothesize--and later confirm--that they will not work …

abstract arxiv box copyright cs.cl cs.cr document fair fair use inference language language models large language large language models llms questions sota train training type