Search papers, labs, and topics across Lattice.
1
0
3
LLMs face a Scale-Dependent Safety Divergence: larger reasoning models achieve safety by adding nuance, whereas capacity-constrained models revert to deleting text.
