#multimodal learning

New research introduces General-Level and General-Bench to measure true synergy in multimodal AI models, revealing current systems lack full integration across tasks and modalities.

Meta AI introduces Web-SSL, a family of large-scale visual self-supervised models trained without language supervision. These models achieve competitive results on multimodal benchmarks, challenging the need for language in vision learning.