AI-Generated Text Detection
Overview
The field studies how to identify whether text was generated or revised by language models, and sometimes which model or spans were involved. Methods range from supervised transformers and stylometric features to zero-shot likelihood, perplexity, token-pattern, and hidden-representation tests. Benchmarks increasingly test multiple languages, domains, generators, paraphrases, and mixed human–AI writing rather than only clean human-versus-machine examples. Detectors can perform strongly on familiar data, but distribution shifts, editing, and stylistic differences can cause missed detections or false accusations, leaving reliable use in high-stakes settings unresolved.
Latest papers
171 papers19 papers in the last four weeks, up 12% on the four weeks before. 0.1% of all new papers.