A new research paper explores the effectiveness of multi-modal large language models (LLMs) in detecting deepfakes. The study benchmarks 12 LLMs, including models like OpenAI O1/4o and Gemini thinking Flash 2, against traditional methods. While some advanced LLMs showed competitive performance, even surpassing traditional methods on out-of-distribution datasets, others performed poorly. The research suggests that while model size can be beneficial, newer versions and enhanced reasoning capabilities do not necessarily improve performance on this specific task, highlighting the potential for multi-modal reasoning in future deepfake detection frameworks. AI
IMPACT Investigates the potential of LLMs for deepfake detection, offering insights into model interpretability and robustness.
RANK_REASON Research paper published on arXiv exploring LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]
- Claude 3.5/3.7 sonnet
- Deepseek Janus
- Gemini thinking Flash 2
- Grok 3
- Llama~3.2
- Mistral Pixtral
- OpenAI O1/4o
- Qwen 2/2.5 VL
- Simiao Ren
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
