What an AI model can and cannot know
A model generates from learned patterns; it does not automatically know private, complete, or current facts.
Understand the system before using it
Durable concepts about model knowledge, context, grounding, data, cost, and risk.
Context versus model knowledge
Task context is evidence supplied for this use; model knowledge is an uncertain capability learned earlier.
Hallucinations and unsupported claims
Unsupported output can sound confident, specific, and internally consistent.
Current facts and source checking
A fluent answer is not evidence that a fact is current.
Grounding and retrieval
Grounding connects an answer to identifiable evidence; retrieval selects evidence before generation.
Data sensitivity and approved environments
Classify data before selecting a model, connector, or storage path.
Inherent task risk versus safeguards
Inherent risk describes the task before controls; safeguards do not erase that underlying consequence.
Cost per completed task
Model price is only one part of cost; review, retries, failures, integration, and escalation also matter.