Search papers, labs, and topics across Lattice.
North China Electric Power University
2
0
5
0
LLM agents are alarmingly susceptible to memory poisoning via conversational attacks, achieving 95% success rates even against agents with selective memory mechanisms.
LLMs can drastically improve entity resolution accuracy, but only if you intelligently integrate them into a global graph refinement process that minimizes query costs.
