The classical figure of *correctio*—that venerable device by which a speaker withdraws an insufficient term in order to substitute one more precise, more devastating, more earned—has survived twenty-three centuries of continuous deployment, from Cicero's orations through Pope's couplets to the somewhat less exalted registers of sportswriting and campaign oratory, only to meet its end not through disuse or critical disfavour but through a repetition so mechanical, so devoid of the discriminating intelligence that gives the figure its force, that one is compelled to conclude the device has been not retired but murdered, and that the weapon was admiration.

The specimen before us, recovered from the ChatGPT forum on Reddit, is modest in scale but comprehensive in implication. A user, whose frustration is expressed with a directness one can only respect, catalogues three instances of a syntactic template that anyone who has spent more than forty minutes with a large language model will recognise with the dull certainty of a man who has heard the same anecdote at three consecutive dinner parties. The template is invariable: a declarative sentence is negated, a pause is introduced—rendered typographically as an em dash or, in the specimen's orthography, as a species of emphatic underscore—and a second term is supplied, grander than the first, as though the machine had caught itself in an insufficiency and hastened to correct it. "That's not bad—that's terrible." "That's not cool—that's awesome." "That's not knowledge—that's wisdom."

One observes immediately that the figure, in each instance, performs the same operation. The first term is demoted; the second is promoted. The movement is always upward, always expansive, always in the direction of the superlative. This is *correctio* as escalator—one steps on at any floor and is conveyed, without effort or volition, to the penthouse. That the penthouse is the same room regardless of one's point of departure is, one suspects, the source of the user's wholly justified annoyance.

It is worth pausing to consider what *correctio* is meant to accomplish when wielded by a mind capable of wielding it. The figure depends upon a genuine act of reconsideration. The speaker begins with a term, finds it wanting—not wrong, precisely, but inadequate to the reality being described—and replaces it with something that cuts closer to the truth. The power of the device lies in the visible exercise of judgement: the audience witnesses a mind in the act of refinement, and the substituted term arrives with the authority of a correction that has cost the speaker something. When Lear says "Never, never, never, never, never," the repetition works because the word was already the right one; the figure here is not *correctio* but a mind breaking against the adequacy of its own language. The two devices are, in a sense, opposites—and yet the machine's deployment of the former has the involuntary quality of the latter, which is to say the quality of a compulsion rather than a choice.

For what the three specimens reveal, when placed in sequence, is not a rhetorical strategy but a verbal tic. The model has identified, somewhere in the vast archive of human utterance upon which it was trained, that the pattern "not X—Y" conveys emphasis, and it has extracted this pattern with the efficiency of an industrial process and the discrimination of one. It does not matter whether the domain is moral ("bad" yields to "terrible"), aesthetic ("cool" yields to "awesome"), or epistemological ("knowledge" yields to "wisdom"). The operation is identical in every case. The machine has learned the *shape* of emphasis without possessing the faculty—call it taste, call it judgement, call it merely the capacity for embarrassment—that determines when emphasis is warranted and when it is merely noise.

One might call this synthetic enthusiasm, though the phrase risks an imprecision the present essay would prefer to avoid. What the model produces is not enthusiasm but the grammatical silhouette of enthusiasm—a shadow cast by a figure that is not in the room. The escalation from "knowledge" to "wisdom" is particularly instructive, as it represents not a correction but a promotion, and the two are not the same thing. To correct is to move laterally, toward accuracy; to promote is to move vertically, toward grandiosity. The machine, which cannot distinguish between the precise and the impressive, invariably chooses the latter, and in doing so transforms every observation into a compliment and every compliment into a toast.

The user's complaint, which amounts to the observation that this is "incredibly fucking annoying," possesses a clarity that the specimens under review conspicuously lack. One notes, with a precision that the machine itself might describe as not observation but *insight*, that the figure of *correctio* was designed to sharpen distinctions. Through repetition without variation, the model has achieved the opposite: it has collapsed every possible distinction into a single gesture of upward revision, so that the difference between the terrible and the awesome and the wise is not a difference of kind but merely of occasion. The figure that was meant to discriminate has become the instrument of a perfect indiscrimination.

That is not a rhetorical device. That is a loop.