Search papers, labs, and topics across Lattice.
TikTok
2
0
3
In-context policy guardrailing remains a daunting challenge, with top models failing to identify even half of the safety violations in straightforward scenarios.
A systematic diagnostic approach reveals that targeted interventions can significantly enhance AVLM performance in video moderation, moving beyond mere trial-and-error fixes.
