Reproduce “TypedBench: A Benchmark for Calibration, Framing Sensitivity, and Cost in System One Decision 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
Rahul Sharma, Andrew B. Duncan, Gaétan Marceau Caron, Sebastian J. Vollmer
TypedBench evaluates typed probability models across wording, calibration, robustness, and decision rules. Seven generators create policy-labelled items, and nine suites measure these behaviors. On 500 generated ticket-priority items, Jev’s accuracy ranged from 0.708 to 0.946 across friendly wordings. Under one specified cost rule, its full probabilities cost more than its top answer; this result is limited to that synthetic task and setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.