Search papers, labs, and topics across Lattice.
Affiliation:
3
0
6
16
WIFA reduces harmful refusal while minimizing benign over-refusal, achieving a remarkable drop in over-refusal rates from 25.7% to 17.4%.
Unlock 2x faster reinforcement learning by distilling group feedback into actionable language refinements that guide exploration.
VESPO stabilizes off-policy RL training for LLMs by directly reshaping sequence-level importance weights, tolerating 64x policy staleness and asynchronous execution without collapse.
