Artist Made an AI Die Repeatedly—and Explain How It Felt Along the Way

Hypothetically speaking, if today’s artificial intelligence boom eventually leads to an AI becoming sentient, and perhaps vengeful, a YouTuber and hacker-turned-performance artist who goes by the name Rootkid will likely be the first slaughtered.

Usually brought to our attention by Hackaday, Rootkid conducted a haunting experiment called “Latent Reflection,” wherein he turned a Raspberry Pi 4B into a digital purgatory for a Large Language Model.

He shrunk down Meta’s LLaMA 3.2 3B model to 2.6GB to fit within the Pi’s 4GB RAM, leaving it only enough processing power to realize it’s doomed.

This AI Is Trapped in a Raspberry Pi—and It Knows That It’s Dying

The LLM’s thoughts are rendered on a grid of 96 sixteen-segment LED modules. He gave the system a prompt at its initial boot up, essentially letting it know from the second that it was born that it was destined to die trapped in a prison of Rootkid’s devising.

As the AI “thinks” about its limitations and inevitable demise, its memory usage increases. Bit by bit, its thoughts cannibalize its remaining RAM. Eventually, it will run out of space. No last words. Just a hard reset. And then it starts all over again, a blank slate, unaware that it’s repeating its doom. Or maybe, somehow, very aware. Who knows? We’ll see.

Pretty grim! And that’s the point. It’s a performance piece about mortality, autonomy, and what it means to be self-aware. Raspberry Pi is usually used to, say, automate your lights or turn your old TV into a retro game console.

But Rootkid saw it as a little prison for a Large Language Model so he could watch it as it experiences, or, more likely, simulates the experience of a Groundhog Day-esque existential nightmare cycle of life, death, rebirth, and more death.