G3Text analysisHallucinationLayer 1 (Deterministic)

Fact Hallucination

Detects factual claims that appear fabricated, such as wrong dates for well-known events, incorrect attributions, or scientific facts that contradict established knowledge.

Technical description

Identifies factual assertions in text (dates of events, named attributions, scientific claims, historical facts) and checks them against a curated knowledge base of commonly hallucinated facts. Focuses on: wrong attribution of discoveries/theories to researchers, incorrect dates for well-known scientific milestones, contradictions with established scientific consensus, and implausible quantitative claims about well-documented phenomena.

How it works

Layer 1 (deterministic): Extracts factual claims (attributions, dates, quantitative assertions) using pattern matching. Checks claims against a knowledge base of commonly hallucinated facts. Validates dates against known timelines. Cross-references attributions against a database of discoveries and their actual originators. Flags claims that contradict established facts.

Why this matters

AI models can state incorrect facts with the same confidence as correct ones, creating convincing but wrong claims. In academic writing, factual hallucinations about well-established science can undermine a paper's credibility and mislead readers. Common examples include attributing discoveries to wrong scientists or citing incorrect dates for landmark studies.

Score thresholds

0-1: No factual errors detected against knowledge base
2-3: Some claims could not be verified
4-5: Multiple factual claims contradict established knowledge

Limitations

The knowledge base is necessarily incomplete and focuses on commonly hallucinated facts. Emerging or contested scientific findings may be flagged incorrectly. Domain-specific facts outside the knowledge base cannot be verified.