Can DreamLLM Surpass the 30% Turing Test Requirement?

DreamLLM, a new multimodal AI system, can generate interleaved documents with text and images based on user instructions. It outperforms existing models in image quality, achieving an FID score of 36.62 compared to 74.77 for Stable Diffusion. In human evaluations, DreamLLM-generated documents received a 60.68% supportive rate, surpassing the 30% Turing test threshold. The system can autonomously create and place images within text, offering a more user-friendly experience than previous models that required manual input for image placement.