Where it fails, and how.
STATING FACTS RELIABLY
Language models generate plausible text. Plausible and true are different things.
They produce confident, well-formed, incorrect statements. That is the central limitation.
ARITHMETIC AND PRECISE CALCULATION
Improving, and still unreliable. Use a calculator or a spreadsheet.
KNOWING WHAT THEY DO NOT KNOW
A system with no relevant information generates something anyway.
It cannot distinguish between recalling and inventing.
CURRENT INFORMATION
Models are trained to a point in time. Anything after that is unknown unless the system can search.
JUDGEMENT IN NOVEL SITUATIONS
Patterns from training do not transfer reliably to genuinely new circumstances.
UNDERSTANDING CONTEXT YOU DID NOT PROVIDE
It knows only what is in front of it.
CONSISTENCY
The same question can produce different answers.
WHAT THIS MEANS
Verify anything factual. Use AI for drafting and thinking, not as a source of truth.
