← Back to Research
Academic Research

FEVER: Fact Verification Dataset

The foundation for ReasonKit's ProofGuard verification system

Authors
Thorne et al. (2018)
Venue
NAACL 2018
ReasonKit Tool
ProofGuard

Paper Title

"FEVER: a Large-scale Dataset for Fact Extraction and VERification"

Key Findings

  • Created 185,445 claims with evidence from Wikipedia
  • Established methodology for fact verification requiring multiple sources
  • Demonstrated that single-source verification fails 40%+ of the time

Why This Matters

Most AI systems trust claims from a single source. FEVER research demonstrates that fact verification requires multiple independent sources to be reliable.

ReasonKit's ProofGuard tool implements this exact methodology, requiring 3 independent sources minimum before trusting any claim—exactly what FEVER research proves is necessary.

Access the Research

View on arXiv (PDF) FEVER Dataset Website

Related Research

Tree of Thoughts (NeurIPS 2023)

Structured reasoning for deliberate problem solving

Self-Refine & Constitutional AI

Iterative refinement with self-feedback - foundation for BrutalHonesty

90% Startup Failure Rate Analysis

Why fact-checking assumptions prevents costly failures

← View All Research Sources