DetectRL

New zero-shot method uses implicit reward models to detect LLM-generated text

Researchers have introduced IRM, a new zero-shot method for detecting text generated by large language models. This approach utilizes implicit reward models derived from publicly available instruction-tuned and base mod…