Bias
1 min read · Reviewed 17 Oct 2026
In short
Skew a model inherited from the material it learned from.
In plain English
A model reflects the patterns in its training data, including the unrepresentative ones. If the material over-represented certain assumptions, so will the output — quietly, and in fluent prose.
A real example
Ask for a description of a nurse, a builder and a chief executive and notice which pronouns and which countries appear by default. Nobody programmed that; it came from the corpus.
Why it matters
It matters most where output touches people: hiring copy, customer segmentation, imagery. The fix is not a better prompt but a human who knows what to look for.
