New UMUI framework enables AI models to make calibrated probability predictions across text, audio, and video modalities simultaneously.

Unified Multimodal Uncertain Inference (UMUI) extends probabilistic reasoning beyond text to audio, video, and multimodal combinations

Researchers created human-annotated evaluation dataset with scalar probability judgments across audio, visual, and audiovisual settings

CLUE (Calibrated Latent Uncertainty Estimation) model uses self-consistent teacher calibration and distribution-based confidence probing for accurate predictions

3B-parameter model achieves competitive performance on existing text and audio benchmarks while handling fine-grained probabilistic reasoning