New TRACE method detects LLM ghostwriters in long texts

Researchers have developed a new method called TRACE to detect ghostwriters generated by large language models in long-form texts. This technique creates a unique fingerprint by analyzing token-level transition patterns, such as word rank, using a separate lightweight language model. TRACE has demonstrated state-of-the-art performance on a new dataset called GhostWriteBench, which includes texts over 50,000 words generated by frontier LLMs, and shows robustness in out-of-distribution scenarios and with limited training data. AI

IMPACT Provides a new tool for identifying AI-generated content in long-form writing, impacting content authenticity and copyright.

RANK_REASON The cluster contains an academic paper detailing a new method and dataset for LLM authorship attribution. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• Anudeex Shetty
• GhostWriteBench
• LLM
• TRACE

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →