Reproduce “Extending Reasoning in Typed Decision Models via Looping” 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

David Berghaus, Arber Zela, Behzad Shomali, Markus Frey, Mehdi Ali, Rafet Sifa, Volkan Cevher

RELOAD lets typed decision models spend extra computation on fixed-choice reasoning by repeatedly applying a middle block to refine hidden representations without generating text. They train it across pass counts. On StepGame stories with 6–10 statements, Von accuracy rises from 70.6% to 85.7% and Brier score falls from 0.404 to 0.198; gains are task-specific, not generalization to arbitrary long reasoning.

Summary by alphaXiv.

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