arXiv Computation and Language

Message capacity and claim wording set the transition points of collective truth-finding in language-model networks

The study investigates how limited reading capacity and claim wording influence consensus outcomes in language‑model networks. By modeling message capacity as the number of messages an agent reads, the authors show that when agents read fewer than about 6.4 messages on average, a wrong consensus becomes unreachable. However, the wording of a claim—its inherent threshold—can override this effect, leading to incorrect consensus even when most agents start correct.

arXiv AI
Oct 31

Network Topology Does Not Matter: The Breakdown of Classical Consensus in LLM Populations

The paper shows that classical consensus theory, which relies on network topology and convex update rules, fails for large language models (LLMs). By measuring token‑level updates across eight instruction‑tuned models and 73,080 exchanges, the authors find that LLMs violate convexity, exhibit negative weights, and are driven by internal update bias rather than network structure. Simulations with 10,000 agents confirm that algebraic connectivity explains almost none of the outcome variance, leading populations to converge on confidently false claims regardless of their network topology.

By Sahil Rajesh Dhayalkar
arXiv Machine Learning
Sep 29

Reinforcement Learning of Communication in a Mesh of Small Language Models

The paper introduces TalkMesh, a decentralized network of small language model agents that learn to communicate effectively during inference. Each agent proposes an answer, scores it with a confidence head, and the most confident agent broadcasts a hint; lower‑confidence agents revise their proposals if a new suggestion scores higher. This gossip‑based consensus, trained via group relative policy optimization, enables a mesh of three agents to match the accuracy of majority voting over 32 samples, and scales to larger meshes to significantly boost performance on benchmarks like GSM8K and MATH-500.

By Mehmet Kerem Turkcan
arXiv AI
Nov 6

Same Behavior, Different Mechanism: What Finite Experiments Can Certify About Integrated Information

The paper investigates when the integrated information measure ζ (phi) can be reliably identified from finite, noisy experiments in small stochastic networks. It shows that even identical input‑output behavior can hide large differences in phi, that partial recordings impose hard precision limits, and that phi is highly sensitive to infinitesimal mechanism changes, often misreported by surrogate models. The authors prove that a lower bound on phi can be certified with controlled experiments and quantify the sample size needed for statistical confidence.

By Linqian Wu, Diaoxuan He
arXiv AI
3d ago

Shared Item-Level Error Limits the Aggregation of Small Language Models in Fermi Estimation

The study examines how shared item‑level error limits the benefit of aggregating small language models in Fermi estimation tasks. By decomposing log10 errors into item‑specific and residual components across 2.51 million answers from 15 open‑weight models, it finds that item‑level error accounts for about 91% of variance, reducing the expected error gain from 74–79% to only 1.7–5.4% when using teams of thirty. Additional techniques such as debate, model mixing, and temperature adjustments offer modest improvements, but no method surpasses the most accurate single model.

By Carolina Fortuna, Blaz Bertalanic
Hugging Face Trending Papers
5d ago

Shared Item-Level Error Limits the Aggregation of Small Language Models in Fermi Estimation

The study examines how shared item‑level error limits the benefit of aggregating small language models in Fermi estimation tasks. By decomposing log10 errors into item‑specific and residual components across 2.51 M answers from 15 open‑weight models, it finds that item‑level error accounts for about 91 % of variance, reducing the expected error gain from a 30‑model team to only 1.7 % (max 5.4 %) instead of the theoretical 74‑79 %. Techniques such as three rounds of debate, model mixing, and temperature adjustments yield modest improvements, but no method surpasses the most accurate single model.