Search papers, labs, and topics across Lattice.
Hong Kong University of Science and Technology (Guangzhou), China
2
0
4
1
Scaling up LLMs boosts combinatorial creativity, but hits diminishing returns for exploratory tasks, revealing a "convergence-by-scaling" effect where larger models become less divergent.
LLMs still struggle with social intelligence, as shown by a new multimodal benchmark revealing limitations in their ability to reason about mental states in multi-turn conversations.
