Reproduce “Sketchpad: certified exact arithmetic with a small vision-language model” 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
Mostafa 'Neo' Mohsenvand
Sketchpad uses a small vision-language model to emit each arithmetic result digit from a three-cell sheet view, with a runtime moving the view. All 2,446 listed views pass on the deployment path; Sketchpad answers 220 of 220 benchmark problems, versus 217 for GPT-6.1 Sol. This guarantee requires certified views and correct runtime arithmetic, not surrounding code or interpretation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.