← Experiments

002·From 2019

Machine

A language model that writes pictures

For the opera I am Alan Turing: portraits were turned into pictures made of text (ASCII) and used to train an instance of GPT-2, which then wrote pictures of its own.

A fine-tuned GPT-3 was later asked for stage directions — “What follows are stage directions narrated by Jorge Luis Borges about Alan Turing’s studio:” — and the text that came back was treated as a starting point for the design, to be answered visually.

Two portraits of a man drawn entirely in rows of text characters: the pictures the model was trained onA head-and-shoulders shape made of rows of text characters on black: what the model wrote

Thread

Machines as collaborators

A language model treated as a generative collaborator: asked for material, which the design then answers by hand.

In the productions: I am Alan Turing