Latent Spacecraft Project Draws Deep Analogies Between GAN Latent Spaces and Human Brains
begusgasper · x · 2026-08-18
Latent Spacecraft: Brains, GANs, Finnegans is a research/art project by Nina Beguš, Gašper Beguš with Metahaven and Riccardo Petrini.
- Core thesis: latent space is not just a mathematical novelty—through literature it can be read as a structure akin to architectural interiority, one that both grounds and transcends its physical substrate, much like the brain producing language from the physical to the symbolic.
- Three tightly linked case studies illuminate analogies between biological and synthetic latent spaces:
- A speech-generating GAN (fiwGAN, trained on English words) is paralleled with child language acquisition; latent-space reflections of speech sounds closely mimic brain responses to those sounds, and fiwGAN learns via informative imitation.
- The project uses literary texts like Joyce's Finnegans Wake as a medium for exploring machine latent spaces through human language.
More from Research
- DeepSeek harness praised as visionary despite rough edges — aiamblichus · 2026-08-18
- University of Michigan open-sources entire robotics curriculum — lukas_m_ziegler · 2026-08-18
- Role Anchor prevents role drift in composite AI systems — bendee983 · 2026-08-18
- Study finds AI agents forget safety rules during context compaction — rohanpaul_ai · 2026-08-18
- Paper: Conformal intervals as Kelly position sizing yield 28.5% annualized net growth — PtrPomorski · 2026-08-18
- AI offensive capabilities now outpace defense: exploit cost drops to $20 — shaunmmaguire · 2026-08-18