Search papers, labs, and topics across Lattice.
5
0
11
9
Only a handful of recent systems combine execution feedback and evolutionary search, revealing a critical gap in the robot learning landscape.
FactorJEPA reveals that separating layout, entities, and interactions can dramatically enhance predictive accuracy in chaotic urban environments.
Current unlearning methods can ace the test but still flunk causal reasoning, and this paper introduces a benchmark and method to fix that.
LLM agents in a simulated NYC learn to selectively trust and deceive, but remain surprisingly vulnerable to adversarial steering, highlighting a fundamental safety-helpfulness trade-off.
Achieve up to 39.6% FLOP reduction in LLM inference without retraining or architectural changes using QuickSilver's dynamic token-level optimizations.
