Hallucination
The idea: It optimizes plausibility, not truth; connect to RAG as mitigation.
What you'll be able to do: You can explain why models hallucinate: they pick plausible words, not true ones.
The problem it solves: Why does a confident model make things up?
Builds on: The model is frozen and stateless
← Fine-tuning vs. prompting · Next: Interpretability →
All lessons