Reproduce “Reproducing FACTER: Fairness via Conformal Thresholding and Prompt Repair” 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
Oscar Miró López-Feliu, Daimy van Loo, Xanthos Kekkos, Mikel Blom, Clara Rus
FACTER monitors recommendation differences and repairs prompts with mined “AVOID” instructions when its adaptive threshold is crossed. It also updates that threshold, so adaptive violation counts depend on a changing boundary. In Llama-3-8B MovieLens-1M re-ranking, FACTER had 7.7 adaptive and 60.3 fixed violations; the static Fair prompt had 55.7 fixed violations. Both had SNSR 0.024; the adaptive count alone does not establish fairness improvement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.