Towards General Conceptual Model Editing via Adversarial Representation Engineering

arXiv:2404.13752v1 Announce Type: new
Abstract: Recent research has introduced Representation Engineering (RepE) as a promising approach for understanding complex inner workings of large-scale models like Large Language Models (LLMs). However, finding practical and efficient methods to apply these representations for general and flexible model editing remains an open problem. Inspired by the Generative Adversarial Network (GAN) framework, we introduce a novel approach called Adversarial Representation Engineering (ARE). This method leverages RepE by using a representation sensor to guide the editing …

adversarial arxiv cs.ai cs.cl cs.cr cs.lg editing engineering general math.oc representation type via