What is Next Token Prediction (NTP) as a core mechanism of generative AI?	Writing answers word by word based on the statistical likelihood of what tends to follow what (sophisticated autocomplete).
What two opposing properties does Next Token Prediction give generative AI?	Fluency and hallucination.
What is the "Capability Zone" on the Next Token Prediction continuum?	Tasks that resemble patterns the model has seen many times (e.g., summarizing or explaining common concepts).
Where does fabrication concentrate when using generative AI?	In specificity, such as names, dates, statistics, citations, and URLs.
What is the primary function of product features like citations or constrained generation?	To mitigate the limitation of Next Token Prediction and push the boundaries of reliable output.
Name four product features designed to help users verify AI output.	Citations, uncertainty signaling, constrained generation, and generator-verifier loops.
Why is the "Verification Test" crucial for users?	Because the same generative process that creates fluency is also the one that causes fabrication.
What does the "Limitation Zone" represent in AI task execution?	Novel or sparse territory, or any task requiring the distinction between "true" and "sounds true."
What does the concept of "Next Token Prediction" explain?	How the AI model generates text output.
What does the concept of "Knowledge" (in this context) address?	What the model actually knows, where that knowledge comes from, and where the knowledge gaps are.