Synthetic Categorical Restructuring large Or How AIs Gradually Extract Efficient Regularities from Their Experience of the World
Michael Pichat, William Pogrund, Paloma Pichat, Armanouche Gasparian, Samuel Demarchi, Martin Corbet, Alois Georgeon, Theo Dasilva, Michael Veillet-Guillem
TL;DR
This study investigates synthetic categorical restructuring in GPT-2XL, formalizing three math-cognitive factors—categorical priming (x-effect), attentional weights (w-effect), and categorical phasing (Σ-effect)—that drive synthetic clipping when building new categories across neural layers, as captured by the aggregation $\sum(w_{i,j} x_{i,j}) + b$. Using a two-layer GPT-2XL setup and a neuron-viewer visualization, the authors quantify how clipped sub-dimensions from precursor categories form new, more efficient target categories, revealing partial confluence, activational dispersion, and categorical distancing. They situate their epistemology in constructivist and embodied cognition and demonstrate, through quantitative and qualitative analyses, that these synthetic processes yield functionally useful restructurings rather than static, preexisting interpretations. The findings illuminate how AI systems actively construct regularities in their token-world experiences, enabling progressively more discriminative representations with potential implications for interpretable AI and cognitive-inspired modeling of language understanding.
Abstract
How do language models segment their internal experience of the world of words to progressively learn to interact with it more efficiently? This study in the neuropsychology of artificial intelligence investigates the phenomenon of synthetic categorical restructuring, a process through which each successive perceptron neural layer abstracts and combines relevant categorical sub-dimensions from the thought categories of its previous layer. This process shapes new, even more efficient categories for analyzing and processing the synthetic system's own experience of the linguistic external world to which it is exposed. Our genetic neuron viewer, associated with this study, allows visualization of the synthetic categorical restructuring phenomenon occurring during the transition from perceptron layer 0 to 1 in GPT2-XL.