Reproduce “VisCurate: Evaluating Behavioral Equivalence of Visual Skills” 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
Md Farhan Ishraq, Md Farhan Ishmam, Reaz Hassan Joarder, Md Hasanul Kabir
VisCurate compares executable image skills by running them on shared probes at matched parameter settings and aggregating output differences by the worst case. In two pilgram versions, 23 of 25 shared-name filters were byte-identical on 177 probes. On 300 corrupted-library instances, it found 51.2% of mergeable pairs at 1.000 precision, but missed 1,614; its certificates cover probes, not every input.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.