This paper reveals cross-modal emergent abilities in multimodal large models — when model scale reaches a critical point, representations from different modalities suddenly fuse, producing cognitive capabilities never exhibited by a single modality alone.
From Isolated Islands to Bridges
Traditional AI systems are like patients with dissociated senses. The CLIP model (2021) first demonstrated the power of large-scale image-text pair training: aligning an image encoder and a text encoder in a shared embedding space.
Deep Cross-Modal Fusion
Imagine not only seeing a vase, but also hearing the designer explain it and touching its glazed texture.
- Shallow multimodality: you see a geometric vase, hear about postmodernism, feel a smooth surface — three independent labels.
- Deep fusion: you understand that the irregular geometry echoes industrial-era fragmented narratives, that the warm texture softens the sharpness, forming a dialectical unity of hardness and softness.
- Google DeepMind & MIT (2025). Cross-Modal Emergent Abilities.
- Radford et al. (2021). CLIP.
Entangled Representation Spaces
The technical core lies in a modality-agnostic concept encoder (MACE). The new architecture adopts early fusion and deep interaction, using cross-modal attention mechanisms. When the model exceeds 100B parameters, reconstruction fidelity suddenly jumps, indicating that an abstract concept layer has formed.
Transfer of Fluid Intelligence
The most striking finding is the cross-modal transfer of fluid intelligence — abstract reasoning ability trained in one modality can transfer zero-shot to another modality.