Armanouche Gasparian

Michael Pichat, William Pogrund, Paloma Pichat et al.

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.

Michael Pichat, William Pogrund, Paloma Pichat et al.

The polysemantic nature of synthetic neurons in artificial intelligence language models is currently understood as the result of a necessary superposition of distributed features within the latent space. We propose an alternative approach, geometrically defining a neuron in layer n as a categorical vector space with a non-orthogonal basis, composed of categorical sub-dimensions extracted from preceding neurons in layer n-1. This categorical vector space is structured by the activation space of each neuron and enables, via an intra-neuronal attention process, the identification and utilization of a critical categorical zone for the efficiency of the language model - more homogeneous and located at the intersection of these different categorical sub-dimensions.

Michael Pichat, William Pogrund, Paloma Pichat et al.

This study investigates the ability of perceptron-type neurons in language models to perform intra-neuronal attention; that is, to identify different homogeneous categorical segments within the synthetic thought category they encode, based on a segmentation of specific activation zones for the tokens to which they are particularly responsive. The objective of this work is therefore to determine to what extent formal neurons can establish a homomorphic relationship between activation-based and categorical segmentations. The results suggest the existence of such a relationship, albeit tenuous, only at the level of tokens with very high activation levels. This intra-neuronal attention subsequently enables categorical restructuring processes at the level of neurons in the following layer, thereby contributing to the progressive formation of high-level categorical abstractions.