LLM Hallucination Error

An LLM Hallucination Error is an LLM error that generates factually incorrect or entirely fabricated information presented as truthful output.

• AKA: LLM Confabulation, Model Hallucination, LLM Fabrication Error, AI Hallucination.
• Context:
  • It can typically generate Nonexistent Facts with confident presentation.
  • It can typically create Fictional Citations appearing authoritative.
  • It can typically produce Imaginary Entities as real references.
  • It can often occur in Knowledge Gaps within training data.
  • It can often increase with Specificity Requests beyond model knowledge.
  • It can often be masked by Fluent Language and plausible structure.
  • It can often mislead Domain Non-Experts through surface credibility.
  • It can range from being a Minor LLM Hallucination Error to being a Major LLM Hallucination Error, depending on its factual deviation.
  • It can range from being a Obvious LLM Hallucination Error to being a Subtle LLM Hallucination Error, depending on its detection difficulty.
  • It can range from being a Factual LLM Hallucination Error to being a Conceptual LLM Hallucination Error, depending on its content type.
  • It can range from being a Isolated LLM Hallucination Error to being a Systematic LLM Hallucination Error, depending on its occurrence pattern.
  • ...
• Example:
  • Citation LLM Hallucination Errors, such as:
    • Inventing Academic Papers with plausible titles.
    • Creating Author Names for nonexistent works.
    • Fabricating Journal References with realistic formats.
  • Entity LLM Hallucination Errors, such as:
    • Generating Fictional People as historical figures.
    • Creating Nonexistent Organizations as real institutions.
  • Fact LLM Hallucination Errors, such as:
    • Inventing Statistical Data without empirical basis.
    • Creating Historical Events that never occurred.
  • ...
• Counter-Example:
  • LLM Conceptual Conflation Error, which confuses real concepts rather than inventing content.
  • LLM Approximation, which estimates uncertain values rather than fabricating facts.
• See: LLM Error, LLM Conceptual Conflation Error, LLM Plausibility Bias, AI System Error, Confabulation, Information Reliability, Fact Checking Task, LLM Factual Accuracy Measure.

---