Search papers, labs, and topics across Lattice.
Affiliation:
4
0
7
AEGIS can effectively defend against Indirect Prompt Injection attacks while maintaining low latency and high utility, a breakthrough for LLM safety.
Adversaries can exploit multimodal guardrails to misclassify safe inputs as unsafe, achieving an alarming 84% success rate in disrupting service availability.
Personal photo retrieval isn't just about visual similarity; PhotoBench reveals that current models fail to leverage the rich context of our lives鈥攖ime, place, people鈥攏eeded to truly understand our search intent.
Current multimodal retrieval systems fall flat when faced with realistic visual streams where context is distributed across time, motivating a new agentic paradigm for context-aware image retrieval.
