A Self-Organized-Model with 3600 neurons, serving as topography for vectorial quests. It was trained on a multi-modal corpus containing more millions of sentences, 3D scans, videos, sounds, poems, materials and artwork.
Onging PhD Project. Artificial Intelligence and Large Language Models offer a glimpse into a veritable meta-verse, a world newly reticulated not only by technical or digital objects but by vectorial hyperobjects that operate in the hyperdimensionality of language. Located far outside the visible, familiar Euclidean space, the world we find ourselves in bestows us with hyper-blindness, starkly opposed to the hyper-vision computer graphics affords us. This vectorial world withdraws from being conventionally analyzed; I propose that while it cannot be queried like a database, it can be quested. This research explores the architectonics of a coded quest as a method to access the vectorial world, drawing on notions from philosophy, philology, literature, and mathematics. Building on a custom-collected, massively multi-modal corpus consisting of thousands of books, 3D scans, sounds, and material entries, it describes how each of these corpuscles can be encoded with constellations of neural nets into words and numbers: each a vector indexing a place in the vectorial world. A SOM as a self-organizing model helps sort these elements in hyperdimensional space. Within a SOM all elements self-organize into a map of the abysmal vectorial world, into which to venture with spectral guides and probabilistic pack animals carrying semantic burdens: epic quests can unfold in unfathomable openness, where novel stories are retold between things and nothings.
The Journey of ALRO_16 in profile and mapped onto the vectorial topography.
Recording of my talk at the Playing Models Conference, November 2023, at the University of Florida.
The course Epic Encounters with online access to corpora and models was taught in the WS 23-24 as CAAD Theory at ETH Zurich.