Search papers, labs, and topics across Lattice.
1
0
2
0
Structural priors can boost semantic vulnerability recall to 100% in synthetic settings but collapse to just 48.9% on real-world data, revealing a critical cross-domain challenge in LLM performance.
