what do you dream about? is an interactive media installation that asks viewers in machine-generated voice questions like "what do you eat for breakfast?", "where did you first fall in love?", "what do you dream about?". Reversing the familiar role of prompting an LLM for our needs, we answer at the behest of the machine, and then it generates in real-time its interpretation of our answer on a giant 100" projection. This generator is conditioned on the camera pointing at the viewer – so audiences begin moving their bodies, turning into strawberries, buildings, and cats. This is done using ChatGPT-augmented prompts and Stream Diffusion, a fast implementation of img2img diffusion that runs at multiple frames per second, but not without jarring hallucinations (if you answer that you dream of running your own business, you will suddenly be looking back at yourself as a man). The machine here feels large, and the viewer small. I brought my setup all over campus, and watched students rock side to side, cover the camera, use props, to condition machine imagination.

The live installation was run on a Runpod RTX 2000 Ada GPU.