Robust Multi-Agent LLMs under Byzantine Faults
arXiv:2605.09076v3 Announce Type: replace-cross Abstract: Large language model (LLM) agents increasingly collaborate over peer-to-peer networks to improve their reliability. However, these same intera...