Search papers, labs, and topics across Lattice.
3
0
3
16
Causal graphs reveal how LLMs perceive concepts, transforming our understanding of model reasoning and interpretability.
LLMs possess domain-specific "gut feelings" about their factual knowledge accuracy that external observers can't see, but this advantage vanishes for math reasoning.
General-purpose agents can match the performance of specialized agents across diverse environments without any environment-specific tuning, challenging the need for task-specific engineering.
