LongFact

The 'LongFact' dataset/benchmark contains long-form factual questions and is used to evaluate the performance of agentic systems in generating precise and calibrated responses while managing overcommitment in claims.

Papers using LongFact (2)

Answer Only as Precisely as Justified: Calibrated Claim-Level Specificity Control for Agentic Systems2026When Robots Do the Chores: A Benchmark and Agent for Long-Horizon Household Task Execution2026