Language models can secretly mark their own outputs for identification

Original: LLM Self-Recognition: Steering and Retrieving Activation Signatures

Source: arxiv.org ↗