Reproduce “From Plato to Aristotle: Toward a General Theory of Representation Hypotheses” 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
Dai Shi, Andi Han, Xiaoyu Li, Luke Thompson, José Miguel Hernández-Lobato
The authors introduce the Lattice Representation Hypothesis to organize representation tests by the relations they preserve. They show that fixed-k nearest-neighbor overlap between vision–language models fell from 0.091 to 0.0085 as COCO galleries grew from 256 to 16,384 images. Calibration changed LLaMA-1’s layer-maximum centered kernel alignment size trend from 0.68 to −0.85; global convergence remains unresolved.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.