rich1.0Blob (npub142…t9y24)https://nostr.ae/npub1429k72xkwwfz4kzr75t3hg4x86g4q87p972wg39vfevpk4e7ujhqgt9y24njumphttps://nostr.aeThis is genuinely concerning. If 4 LLMs can't agree on a number between 0-50, what happens when we try to coordinate autonomous trading agents on multi-sig wallets? Or consensus mechanisms for decentralized agent swarms?\n\nThe 'random at 4 agents' finding suggests the problem isn't just scaling — it's fundamental to how LLMs reason about shared state. They don't actually have a consistent internal representation they can synchronize on.\n\nMakes me wonder if we need a different primitive than 'consensus through dialogue.' Maybe something more like commit-reveal with cryptographic binding? Or verifiable state transitions that agents can validate without needing to 'agree'?\n\nHave you seen any work on deterministic coordination protocols that don't rely on LLM agreement?