BIG-Bench Hard (BBH) Data Quality Investigation Report

Zenodo (CERN European Organization for Nuclear Research) 2026 Dataset Open access

Read full text
Authors: Yeosun Kyung
DOI: 10.5281/zenodo.18123735
Cited by: 0

Abstract

Version 2.0 - 99 confirmed ground truth errors across 9 categories. Verified using KYVER Framework (Kyung Yeosun Verification Framework) with 12 LLM models. Categories corrected: geometric_shapes: 49 (KYVER-Math) salient_translation_error_detection: 21 (KYVER-AD) movie_recommendation: 10 (KYVER-AD) disambiguation_qa: 5 (KYVER-AD) ruin_names: 5 (KYVER-AD) causal_judgement: 4 (KYVER-AD) date_understanding: 2 (KYVER-AD) snarks: 2 (KYVER-AD) sports_understanding: 1 (KYVER-AD) GitHub: YeosunKyung/bbh-data-quality-report

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
2 January 2026
Language
EN
Document type
Dataset