Reproduce “Readable Is Not Steerable: Pre-Registered Tests of Physical Structure in Learned Fluid Representations” 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
Manvendra Modgil
The authors test whether properties are readable, prominent, and controllable in latents of models trained on 64×64 periodic, decaying two-dimensional flow. Real-dynamics JEPA slow directions tracked enstrophy less well than an untrained control, while rendering autoencoders made it prominent. A routed decoder produced valid loudness dials in five seeds but shape dials in only three, limiting conclusions to this setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.