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