
Cet essai part d’une expérience de travail prolongée avec les grands modèles de langage — non leur usage, mais leur construction et l’instrumentation de leurs défaillances — et décrit un retournement inattendu : la représentation, tenue depuis toujours pour l’instance devant laquelle le réel est en défaut, devient porteuse d’une permission de limite. Deux propriétés de ces systèmes la rendent possible. Leur intelligence est un archipel : parfaite par plaques, nulle juste à côté, sans conscience du contraste. Et leur population est ordonnée en une hiérarchie publique et mesurée — petits, moyens, grands modèles — qui rejoue la hiérarchie des capacités cognitives humaines, sujet jusqu’ici impossible à regarder sans blesser, et montre qu’en descendant l’échelle, l’intelligence ne s’anéantit pas. L’image mécaniste de la pensée qui en résulte rend la finitude cognitive dicible comme donnée neutre plutôt que comme disgrâce. L’essai nomme ce déplacement l’humain readymade — une rédemption sans amélioration, au sens de Duchamp — et en cherche la figure dans la scène finale d’A.I. (Kubrick/Spielberg, 2001) et la quasi-conclusion de L’Écriture du désastre de Blanchot : une douceur sans préalable conscient, qui émerge entre deux pauvretés quand aucune n’a plus à se justifier devant l’autre.
This essay draws on an extended experience of building and instrumenting large language models — measuring their failures rather than consuming their outputs — and describes an unexpected reversal: representation, long held to be the instance before which the real stands in default, becomes the bearer of a permission of limit. Two properties of these systems make this possible: their intelligence is an archipelago, perfect in patches and null immediately beside them, with no awareness of the contrast; and their population is ordered into a public, measured hierarchy of small, mid-sized and large models, replaying the hierarchy of human cognitive capacities — a subject hitherto impossible to examine without offense — while showing that intelligence, descending the scale, is not annihilated. The resulting mechanistic image of thought makes cognitive finitude sayable as neutral fact rather than disgrace. The essay names this displacement the readymade human — a redemption without improvement, in Duchamp’s sense — and finds its figure in the final scene of A.I. (Kubrick/Spielberg, 2001) and the closing wager of Blanchot’s The Writing of the Disaster: a gentleness without conscious precondition, emerging between two poverties when neither has to justify itself before the other.
Duchamp, finitude, philosophy of technology, Blanchot, permission de limite, large language models, artificial intelligence, readymade, human-machine interaction
Duchamp, finitude, philosophy of technology, Blanchot, permission de limite, large language models, artificial intelligence, readymade, human-machine interaction
| selected citations These citations are derived from selected sources. This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | 0 | |
| popularity This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network. | Average | |
| influence This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | Average | |
| impulse This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network. | Average |
