Reproduce “Revisiting LLM Homogeneity Through a Multidimensional Lens” 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

Ziliang Qiu, Renfen Hu, Yi Ding, Ted Underwood

The authors measure language-model response homogeneity with 25 interpretable measures across semantic, lexical, syntactic, structural, and stylistic dimensions, using human ratings and comparisons across 18 models, 15 releases, and two post-training pipelines. Across generations, within-model distances fell for 17 of 22 changing metrics, while between-model distances rose for 14 of 21; these descriptive trends do not establish causes.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.