Reproduce “From Interface to Inference: Eliciting Any-Order Inference from Any-Order 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
Seunggeun Kim, Jaeyeon Kim, Taekyun Lee, Yuyuan Chen, Yilun Du, Sham Kakade, Sitan Chen
The paper tests why masked diffusion models with any-position prediction may still generate code much like left-to-right systems. It attributes this to positional uncertainty and studies insertion-based decoding and latent segment selection. On zero-shot GSM8K at parallel decoding size k=1, LatentMDM’s confidence-based selector scores 45.5%, versus 41.3% and 40.3% for alternative policies in the same model, limited to this setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.