Glossary
Metacognitive laziness
Metacognitive laziness is the tendency to accept fluent AI output without evaluating it, and the gradual erosion of the judgement needed to evaluate it at all. The more polished the output, the less it invites scrutiny.
In more detail
The term describes a feedback loop rather than a character flaw. Language model output is fluent by construction — that is what the training optimises — and fluency is one of the strongest heuristics people use to judge quality. Text that reads well is assumed to be well reasoned, and the assumption is usually close enough to be reinforced.
Over time the loop tightens. Each accepted output is one skipped act of evaluation, and evaluation is itself a practised skill. The capacity to judge whether output is any good decays through the same disuse as every other capability, which means the signal that would break the loop degrades alongside it.
This is why IMPRINT's Mirror refuses to answer. A surface that only asks questions cannot be accepted uncritically, because it never supplies anything to accept — the evaluation stays with you by design.