LLM Map-Reduce Pattern NEW

Problem

Injecting a single poisoned document can manipulate global reasoning if all data is processed in one context.

Solution

Adopt a map-reduce workflow:

• Map: Spawn lightweight, sandboxed LLMs—each ingests one untrusted chunk and emits a constrained output (boolean, JSON schema, etc.).
• Reduce: Aggregate those safe summaries with either deterministic code or a privileged LLM that sees only sanitized fields.

results = []
for doc in docs:
    ok = SandboxLLM("Is this an invoice? (yes/no)", doc)
    results.append(ok)
final = reduce(results)  # no raw docs enter this step

How to use it

File triage, product-review summarizers, resume filters—any N-to-1 decision where each item's influence should stay local.

Trade-offs

• Pros: A malicious item can't taint others; scalable parallelism.
• Cons: Requires strict output validation; extra orchestration overhead.

References

• Beurer-Kellner et al., §3.1 (3) LLM Map-Reduce.