Beyond Consensus: Trace-Level Synthesis in Mixture of Agents
Shreyas Fadnavis, Praitayini Kanakaraj, Felix Wyss
Why It Matters
What makes this one worth your time
This approach could enhance the accuracy and robustness of AI systems by utilizing diverse reasoning paths, which is crucial for complex problem-solving tasks.
Aggregating reasoning traces from multiple agents can outperform consensus-based methods.
Summary
The paper proposes a method for aggregating reasoning traces from multiple LLM agents to improve problem-solving accuracy, challenging the traditional consensus-based approaches. It introduces a Self-Consistent Mixture of Agents framework that leverages trace diversity through semantic-preserving input perturbations and anchored refinement to ensure non-degradation of results.
Key contributions
- Introduction of a Self-Consistent Mixture of Agents framework for trace-level synthesis.
- Demonstration that trace diversity can outperform heterogeneous model pools in various domains.
Notable insights
- Trace-level synthesis can recover correct solutions even when agents agree, highlighting the potential of minority reasoning paths.
- Semantic-preserving input perturbations can generate trace diversity without degrading the majority's accuracy.
Possible limitations
- Not stated in the abstract
Abstract
arXiv:2605.29116v1 Announce Type: new Abstract: When multiple LLM agents solve the same problem, standard practice compresses each agent's reasoning into a majority vote or layered synthesis, treating agreement as the finish line. We show this is unnecessarily lossy: an LLM aggregator that reads complete reasoning traces recovers correct solutions even when agents unanimously agree, with beneficial corrections consistently outweighing harmful ones -- the \emph{aggregation paradox}. Majority voting has a ceiling that perturbation diversity does not raise (error correlations are identical); the aggregator's gain comes from trace-level complementarity, assembling correct intermediate steps from minority chains that voting discards. These findings motivate Self-Consistent Mixture of Agents which generates trace diversity through semantic-preserving input perturbations, safeguards the majority via anchored refinement with provable non-degradation guarantees, and always synthesizes -- never gates on consensus. A single model with perturbation-induced trace variation outperforms heterogeneous model pools across structured reasoning, PhD-level science, competition mathematics, and competitive programming. The unit of aggregation should be the reasoning trace, not the answer.