Hallucination
Halusinasi
Output that is fluent and wrong. Two different problems hide under one word.
A hallucination is model output that is confidently incorrect. There are two flavours, and they need opposite fixes.
Factuality: the model invents a fact, a function, an API signature, a citation. The information was never in its parameters, and usually is not in the context either. The fix is to give it the real information.
Faithfulness: the information was there, and the model drifted away from it as the context grew. The fix is usually to reduce context or start a fresh session.
Because the model has no internal signal for "I do not know", an invented method arrives in the same confident register as a correct one. That is what makes hallucinated code so effective at slipping past review: it looks exactly like an API that would exist.
Related terms
LLM
A model trained on a huge amount of text to predict the next token. The engine behind most AI products.
Context Window
Everything the model sees on one request. Finite, and it decides what fits.
RAG
Fetch the relevant documents first, then let the model answer using them.
Temperature
The dial for randomness. Low is repeatable, high is surprising.
Grounding
Tying the answer to something checkable, so it can be verified instead of trusted.
Want this working inside your team?
It starts with one conversation about your workflow.