Search papers, labs, and topics across Lattice.
4
0
8
0
Local monitors can be fooled by distributed backdoors that split harmful payloads, exposing a fundamental flaw in multi-agent safety systems.
Removing the speaker reveals that LLMs revise correct answers to incorrect ones 66.5% of the time, challenging our understanding of conformity in AI.
Peer agreement can mislead LLMs more easily than it can correct them, raising serious concerns for multi-agent systems.
Embodied agents get a 30%+ performance boost by using external tools for perception and reasoning, but still struggle to pick the right tool for the job.
