Reproduce “Behavioral Probes for Information Flow in LLM Swarms” 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
Junhui Chen, Yu Zhang, Lantian Li, Dong Wang
The authors adapt ECS and PKS as heuristic probes of how language-model swarms use external messages and parametric knowledge. ECS measures output-context relationships, while PKS tracks changes in token predictions across late feed-forward blocks. In controlled tests, ECS distinguished domain-matched expert messages, and Probe-Joint reduced redundant inbound weight and tokens per question, with 80.0% accuracy versus GPT Swarm’s 80.6%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.