Safety & Ethics
Hallucination
When a model generates plausible-sounding but factually incorrect information.
Hallucination occurs because LLMs generate the statistically likely next token rather than retrieving verified facts. A model may confidently cite nonexistent papers, invent API signatures, or fabricate statistics. Mitigation strategies include grounding via RAG, tool use for fact-checking, and choosing models with lower hallucination rates on benchmarks like TruthfulQA.
Related Terms
RAG (Retrieval-Augmented Generation)
Enhancing model responses by fetching relevant documents from an external knowledge base at query time.
Grounding
Connecting model responses to verified, real-world information sources.
Tool Use (Function Calling)
The ability for a model to call external functions or APIs to retrieve data or take actions.
Knowledge Cutoff
The date beyond which a model has no training data and is unaware of events.