Search papers, labs, and topics across Lattice.
1
0
3
Multimodal LLMs suffer a major performance hit when asked to switch from text-based to image-based tasks mid-conversation, revealing a surprising asymmetry in their ability to handle task interference.
