Can ChatGPT prevent deepfakes? Study tests LLMs’ abilities #AIsecurity

The study explores the potential of using large language models (LLMs) like ChatGPT and Gemini to detect deepfake images. While LLMs lag behind state-of-the-art deepfake detectors in accuracy, they excel in explaining their analysis in plain language, making them more practical for future detection efforts. The study found that ChatGPT could accurately detect synthetic artifacts in images and explain its decision-making process, providing a user-friendly approach to deepfake detection. However, LLMs struggle with signal-level statistical differences that traditional detection algorithms can catch, limiting their effectiveness. Despite these drawbacks, the study highlights the potential of LLMs in deepfake detection due to their semantic knowledge and natural language processing abilities. The research team emphasizes the need for further fine-tuning LLMs for this specific task to improve their efficiency. Overall, the study suggests that LLMs like ChatGPT could play a significant role in combating deepfake images by leveraging their unique capabilities in semantic understanding and explanation.

Source link

Source link: https://techxplore.com/news/2024-06-chatgpt-key-deepfakes-llms-ai.html