<ul><li>Adversarial attacks are being used to test the robustness of AI-generated text detection models.</li><li>A novel token probability-based approach using embedding models is proposed to reduce the likelihood of detection of AI-generated texts.</li><li>The method utilizes different embedding techniques, including the Tsetlin Machine (TM), to perturb the data and reconstruct the texts.</li><li>The proposed method shows a significant reduction in detection scores against Fast-DetectGPT on XSum and SQuAD datasets.</li></ul>

Adversarial Attacks on AI-Generated Text Detection Models: A Token Probability-Based Approach Using Embeddings

Discover more