CONCEPT
Aliasing and Hallucination
The structural connection between Nyquist's aliasing—the confident false signal produced by insufficient sampling—and the language model's hallucination—the fluent false claim produced by sparse training in a domain—both instances of the same phenomenon: structured falsehood that is indistinguishable from truth because it has stolen the appearance of legitimacy.
Aliasing is
Harry Nyquist's darkest contribution to understanding artificial intelligence. When a signal is sampled too slowly—below the Nyquist rate—the frequencies that cannot be represented do not vanish into silence. They fold back into the sampled data as lower frequencies that were never in the original, indistinguishable from genuine signal, and irrecoverable: once the false structure has been fused with the true, no downstream processing can separate them. The wagon wheel that appears to spin backward in a film is aliased: it is not a blurry or degraded representation of forward rotation, but a confident, sharp, completely false depiction of backward rotation. The large language model that invents a legal precedent, attributes a concept to the wrong philosopher, or connects two ideas in a way that sounds insightful but is factually wrong is producing the linguistic equivalent: not noise, not hedged approximation, but confident structured falsehood wearing the