Reproduce “Coherence-Driven Belief Formation and Population Dynamics of Contagion in LLM Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tathagata Banerjee, Nima Moghaddas
The authors measure how peer endorsement changes LLM agents’ claim acceptance, then use the responses in network simulations. Adoption rises sigmoidally with endorsing peers, while claim plausibility, source reliability, and agent disposition shift its threshold. In a small simulation, clustered networks produced larger cascades in a tested regime, and consensus persisted after pressure ended; these findings concern unverifiable forecasts, not real-world misinformation.
Key insights
- Adoption generally rose in a threshold-like manner with peer endorsement, while plausibility, source reliability, disposition, and stated priors shifted the number of endorsements needed.
- In the tested conditions, these factors were largely captured by a single effective dimension: the first principal component explained 98% of threshold variation in the primary Qwen study.
- Belief spread showed history dependence: conditions that were difficult to establish could remain stable after consensus formed, even as the supporting pressure was reduced.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.