| This year, my project has developed from an interest in live information streams into a system for translating speech and text into architectural form. This led me to question how large language models process language and how their hidden operations might be made visible. The project now investigates LLMs not as tools for producing images, but as probabilistic language systems. Each word or token is read through the model’s confidence: expected words produce one type of spatial behaviour, while uncertain or surprising words generate another. These invisible statistical relationships are translated into colour, geometry and material logic. Through this process, text becomes a spatial trajectory. Token confidence generates form, rhetorical structures produce larger volumes, and sequences of language become branching architectural systems. The resulting geometry is not a representation of what the text means, but a translation of how the model statistically reads it. Architecture therefore emerges through a feedback loop between user and machine. I provide language, the system interprets it probabilistically, and I then reinterpret the output architecturally through form, material and atmosphere. The project proposes an architecture shaped by uncertainty, authorship and the hidden structures of computational language. |
