Search papers, labs, and topics across Lattice.
2
0
6
8
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.
