multi-agent-llm-systems-where-one-agent-starts-erroneous-shared CLM-0098In multi-agent LLM systems where one agent starts with an erroneous shared belief, broadcasting every agent's full context at every step raises the rate of false assertions above doing no synchronization at all, because the error propagates to agents that were previously correct — and the harm appears only in tasks where one wrong fact cascades across semantically linked dimensions (destination to airport to weather), not where agent contexts are largely orthogonal.
Evidence for: Ashby's law of experience (holds), Braess's paradox (holds), Linus's law (breaks), Two heads are better than one (breaks)
Capability: Stating false facts confidently
Observed on
Three-agent planning tasks with deliberately injected context mismatches, Claude Haiku agents; travel-planning domain showed the effect, software sprint planning did not..
Sources
- Measured on 30/30/15 trials in travel and 10/10/5 in software with one model. Full-broadcast HR 0.658 vs no-sync 0.492 (p=0.0022, d=1.18) in travel; effect absent in software. Errors were deliberately injected, so the result isolates propagation of a seeded false belief rather than naturally arising drift. Single author, single model, small n.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- In LLM multi-agent systems that reflect after failure, restricting reflection to the single agent responsible for the first incorrect step — located by a trained step-level correctness scorer and paired with a counterfactually generated corrected step — raises task success more than having every agent reflect on the trajectory, because reflections written by agents that behaved correctly enter their memory as misleading guidance for later tasks.Getting more from several models than from one · reviewed by AI
- In LLM multi-agent systems whose communication graph is produced by a learned topology generator, the generated graph carries edges that are not needed: scoring each edge by how much masking it changes the task outcome and the semantic entropy of the final answer, then keeping a top-ranked budgeted subgraph, cuts online tokens by about 20-25% while leaving average accuracy equal or slightly higher, and a small distilled predictor of those subgraphs transfers to generators and agent counts it was not calibrated on.Getting more from several models than from one · reviewed by AI
- When automatically rewriting the prompts of a multi-agent LLM system from its failures, locating the responsible agent by injecting a hint into one agent at a time and using that agent's hint-corrected intermediate output as the supervision signal produces prompt updates that generalize better than critique derived only from the system-level output.Improving its own scaffolding from failures · reviewed by AI
- In hierarchical LLM agents acting in gridworld cooperation tasks (Overcooked layouts) with unseen partners, supplying an accurate partner-skill estimate as planner prompt context does not by itself stop the agent from continuing a stale, non-complementary skill; the estimate only changes behaviour when it is wired into an explicit interrupt rule that aborts the in-progress skill when new partner actions are unlikely under the committed estimate.Turning plans into physical action · reviewed by AI
- In multi-agent LLM networks where a correct minority peer disagrees with an agreeing majority, ranking peers by raw divergence from the receiver's own belief systematically prefers the wrong peers, while scoring each peer by asserted precision divided by its divergence from a second-order model of what that peer should have concluded given its data subscription isolates the correct one.Getting more from several models than from one · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: Full-context broadcast between agents matching or lowering hallucination rate relative to no synchronization in cascading-belief tasks, across models and injected-error setups. Proposed technique, not catalogued: threshold-gated context synchronization between agents.