What it fundamentally cannot do.
RELIABLE FACTUAL ACCURACY
Not solved. Models generate plausible text, and plausible is not true.
KNOWING WHAT IT DOES NOT KNOW
A model cannot distinguish between recall and invention.
GENUINE UNDERSTANDING
Whether these systems understand anything is contested. Practically, they do not behave like something that does.
They fail in ways a person who understood would not.
REASONING ABOUT NOVEL SITUATIONS
Pattern-matching resembles reasoning and diverges from it in unfamiliar territory.
LEARNING FROM FEW EXAMPLES
Humans learn a concept from one or two examples. These systems require enormous quantities.
CONSISTENCY
The same question may produce different answers.
ACCOUNTABILITY
A system cannot be responsible for a decision. A person must be.
WHAT THIS MEANS
Use AI where its strengths apply and its failure modes are tolerable.
Do not deploy it where being confidently wrong would cause serious harm without human review.
