Reproduce “Worldviews in Large Language Models” 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
Edmundo Lopez-Sola, Shamil Chandaria, Ruben Laukkonen
The paper measures models’ expressed worldviews and tests whether dialogue changes survey responses and safety behavior. Across eight models, it finds scores below a human comparison sample on all four belief scales, with 53 of 119 worldview positions unanimous. Safe responses rose most after dialogues classified as spiritual-bliss, but this association does not establish causation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.