Safety
Hallucination
A phenomenon where LLMs generate plausible-sounding but factually incorrect or fabricated information, often with high confidence.
Explained at five levels
Level 1
When the AI makes something up and says it like it's true, even though it's not — like telling a really convincing fib.
Level 2
When an AI confidently gives you wrong information — it sounds right but it just made it up. That's why you should double-check important facts.
Level 3
A phenomenon where LLMs generate plausible-sounding but factually incorrect or fabricated information, often with high confidence.
Level 4
The generation of content that is syntactically fluent but semantically unfaithful to source data or world knowledge, arising from the model's tendency to optimize for fluency over factual accuracy.
Level 5
Unfaithful generation where the model's output diverges from the grounding context or world model — categorized as intrinsic (contradicting the source) or extrinsic (unverifiable), driven by exposure bias, knowledge conflicts, and decoding dynamics.
Definitions are educational summaries. Terminology can vary by source and context.
Sources
- NIST Trustworthy and Responsible AI Resource Center: Glossary — National Institute of Standards and Technology. Accessed 2026-07-20.