A tendency for a model's outputs to converge on one likely, average answer instead of representing the full spread of real opinion.
Continue to AI University →