Reproduce “ValueFlow: Measuring the Propagation of Value Perturbations in Multi-Agent LLM Systems” 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
Jinnuo Liu, Chuke Liu, Hua Shen
ValueFlow examines how value signals affect responses within interacting language-model systems. It perturbs peer messages, measures individual-agent responsiveness and system-level propagation, and tests value-loaded prompts against tone-only prompts. Across ten values, mean output deviation is 6.12 versus 1.94. A controlled three-agent writing case shows framing persists downstream; these behavioral findings do not establish latent values or universal effects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.