Short Form Factual Question Dataset Curation

via “short-form-factual-question-dataset-curation”

OpenAI's factuality benchmark for hallucination detection.

Unique: Explicitly curates for short-form questions with unambiguous answers to isolate factuality measurement, rather than using general QA datasets that mix factuality with reasoning, comprehension, and inference complexity

vs others: Cleaner factuality signal than general QA benchmarks because it removes confounding variables like reasoning complexity, enabling precise attribution of errors to hallucination rather than reasoning failures