Rethinking Design Fixation in LLMs: A Process-Based Perspective Toward Linguistic Fixation

Marco Consoloni, Antoine Bordas (), Filippo Chiarello, Gualtiero Fantoni and Pascal Le Masson ()
Additional contact information
Marco Consoloni: UniPi - University of Pisa [Italy] = Università di Pisa [Italia] = Université de Pise [Italie]
Antoine Bordas: CGS i3 - Centre de Gestion Scientifique i3 - Mines Paris - PSL (École nationale supérieure des mines de Paris) - PSL - Université Paris Sciences et Lettres - I3 - Institut interdisciplinaire de l’innovation - CNRS - Centre National de la Recherche Scientifique, Mines Paris - PSL (École nationale supérieure des mines de Paris) - PSL - Université Paris Sciences et Lettres
Filippo Chiarello: UniPi - University of Pisa [Italy] = Università di Pisa [Italia] = Université de Pise [Italie]
Gualtiero Fantoni: UniPi - University of Pisa [Italy] = Università di Pisa [Italia] = Université de Pise [Italie]
Pascal Le Masson: CGS i3 - Centre de Gestion Scientifique i3 - Mines Paris - PSL (École nationale supérieure des mines de Paris) - PSL - Université Paris Sciences et Lettres - I3 - Institut interdisciplinaire de l’innovation - CNRS - Centre National de la Recherche Scientifique

Post-Print from HAL

Abstract: Large Language Models (LLMs) are increasingly employed in design tasks, raising the need for robust evaluation methods. Existing studies focus on assessing LLM-generated outputs using human-centered design constructs, such as design fixation, overlooking the underlying process through which LLMs generate text. This outcome-based approach neglect that design fixation may stem from the LLMs generation process, and that the "creativity" of LLMs is inherently (upper) bounded by training data. This study investigates design fixation at the process level by examining its origins within LLM architecture. We perform a clustering analysis of hidden states — high-dimensional vectors that LLMs use to predict the next token. Results show that hidden states are primarily organized by linguistic features, such as part-of-speech categories, rather than by design-oriented categories. This suggests that design fixation may emerge as a text-level property attributable to human evaluation of LLM outputs, rather than as a characteristic of the LLM's generative process.

Date: 2026-07-08
Note: View the original document on HAL open archive server: https://hal.science/hal-05591627v1
References: Add references at CitEc
Citations:

Published in Design Computing and Cognition DCC'26, Jul 2026, Paris, France. pp.15-34

Downloads: (external link)
https://hal.science/hal-05591627v1/document (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:hal:journl:hal-05591627

Access Statistics for this paper

More papers in Post-Print from HAL
Bibliographic data for series maintained by CCSD ().