Deep World is a live audiovisual / AI performance art by VJ Übergeek, Amy Alexander’s long-running performance persona. The piece explores the evolving relationship between humans and chatbots—from their early roles as informal therapists to their current status as omnipresent advisers in tumultuous times. As Übergeek confides in Deep World AI—a custom language model trained on writers and thinkers who wrestled with difficult times—the performance unfolds as part live text improv, part VJ set, and part video game. Deep World reflects on the porous boundary between personal outlook and collective conditions, probing how we seek meaning, clarity, and a way forward in moments of crisis—and what it means when the answers come from a machine trained on past attempts to survive the world.
Deep World picks up where Übergeek’s CyberSpaceLand (2000–2015) performance for nightclubs, streets, and art spaces left off. Like CyberSpaceLand, Deep World is a text dialogue between performer and the cyberspace “world,” rolled into a semi-theatrical VJ performance by cyberhuman would-be rock star Übergeek, who drives the performance with game controllers and gestural interfaces. But with the emergence of AI language models (LLMs), the dialogue has shifted: narrative improvisation between performer and machine has become more central to the performance. Alongside public space performance, Übergeek plans some live-streamed internet improvisations for Spring/Fall 2025, as well as possible standalone interactive installations.
Deep World debuted in June 2025, and it’s still under development. The debut performance featured DJ monocromo.
Performance footage courtesy of Memo Akten.
Sample Screenshots:
The Deep World LLM is (currently) based on Llama 3.x running under Ollama, with customizations by Amy Alexander.
The visual performance application and effects are developed by Amy Alexander using Max/MSP/Jitter. A couple of bits are adapted from tutorials by Andrew Benson and Andrew Robinson.
The ASCII text animations use the TerminalTextEffects library.