The Life Cycle of an AI on Display

Some hobbyist electronic projects are meant to solve a problem, and others are just designed to make us think. YouTuber Rootkid’s latest art project, called Latent Reflection, falls into the latter category. Specifically, Latent Reflection invites us to imagine what artificial intelligence (AI) models might be thinking. Of course they cannot actually think at all, but it is fun to imagine what some futuristic, sentient AI might ponder in its free time all the same.

Since the project is first and foremost a work of art, the build process started with the development of a unique display. Rootkid created this display from a 16 x 6 matrix of 16-segment display modules. The matrix was built from a series of custom PCBs, each containing eight display modules, that were daisy-chained together. A Raspberry Pi 4 Model B single-board computer controls the state of the display with the help of a driver written by Rootkid. The components were all mounted to a sheet of aluminum.

To test things out, a video of Rootkid’s eye was recorded, then converted into a two-color version to show on the display. A Python script converted the pixels in each image into segments to show on each module. Finally, the custom display driver played the video in real-time.

With that out of the way, Rootkid moved on to the AI. He first tracked down a large language model (LLM) that could comfortably fit within the resources available to the Raspberry Pi. Specifically, a Llama 3.2 model with three billion parameters, that had been quantized for the best performance on resource-constrained platforms, was selected.

This algorithm was then given a prompt telling it that it was an LLM, where it was running, that it could be turned off at any time, and other such bleak-sounding information. It was then instructed to produce a steady stream of its “thoughts.” The response is shown on the display as it is produced until the computer inevitably runs out of memory and the model crashes, at which point it resets and starts all over.

Even though this is just an algorithm predicting the next most likely word in a sequence, the result feels pretty dark. It is enough to make us feel sorry for this computer algorithm that knows and feels nothing. Mission accomplished, Rootkid. Latent Reflection will definitely make us reflect on AI, life, and other heavy matters.