This set of four Towers of Hanoi forms part of an installation created by Julien Prévieux for his solo exhibition Des raisonnements déraisonnables (Le Grenier à Sel, Avignon, April–June 2026). The installation brings together a collection of errors gathered through experiments involving different chatbots, including ChatGPT, Claude, DeepSeek, Gemini, Mistral and Qwen.
Here, Julien Prévieux gives material form to something that occupies a central place in contemporary artificial intelligence models: attention. At the heart of language models, attention does not refer to the ability to concentrate, but to a statistical weighting operation that can go awry. Chessboards, Towers of Hanoi, large diagrams and distorted texts materialise these multiple failures, made up of impossible sequences and false solutions.
Julien Prévieux transforms computational malfunctions into visual and sculptural objects. Together, they form a record of failed operations: the remnants of a process devoid of any overall understanding, shaped by dead ends and unstable forms of logic. These objects are presented as an endless collection to which it is up to us to pay genuine attention.