Search papers, labs, and topics across Lattice.
3
0
6
5
LLM agents can achieve high task accuracy while exhibiting profound epistemic *im*humility, failing to acknowledge uncertainty even when presented with contradictory evidence.
Programmatic skill learning can slash agent costs while enhancing performance, with SpeedRunner leading the charge in cost-efficient adaptation.
Controlled task streams reveal that naive memory designs fail to leverage reusable knowledge effectively, exposing critical gaps in continual learning evaluations.
