Findings of the Counter Turing Test: AI-Generated Text Detection

In three linesCounter Turing Test evaluates AI-generated text detection techniques. Task A (binary classification) achieves F1=1.0 to distinguish human vs AI text. Task B (model attribution) reaches 0.9531 for identifying GPT-4, Claude 3.5, Llama. Top approaches combine DeBERTa, BART, fine-tuning, and ensemble learning.

Summary generated by Claude — human-verified