Reproduce “Where Does Alignment Emerge and What Does It Mean? A Closer Look at the Platonic Representation Hypothesis” 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
Julia Herbst, Simon Roschmann, Quentin Bouniot, Zeynep Akata
The authors compare separately trained vision and language models to test where their image-text neighborhoods agree. They measure mutual k-nearest-neighbor overlap across layer pairs using image-caption examples. On WIT, agreement is concentrated in later layers and weaker than within-modality agreement. On DOCCI, alignment correlates with some benchmark scores, but not GSM8K; this association does not establish causation or general capability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.