Search papers, labs, and topics across Lattice.
University of Central Florida
3
0
6
Safety-aligned text-to-image models may appear effective, but they suffer substantial semantic fidelity losses that standard metrics fail to capture.
Reasoning models may boost performance but often sacrifice critical alignment behaviors, revealing a hidden trade-off in AI safety.
Optimizing LLMs for generating multiple attempts (pass@k) can actually *hurt* their ability to get it right on the first try (pass@1) due to subtle prompt interference effects.
