Stress-testing AI-text detectors: Binoculars (zero-shot) vs TF-IDF/CNN/RoBERTa on 9,458 TuringBench human texts paired with Llama-2 generations, under decoding shifts and paraphrase attacks
nlp scikit-learn pytorch georgia-tech robustness llm binoculars llama2 ai-text-detection turingbench
-
Updated
Jul 20, 2026 - Python