Where Thought Never Lands

Single-channel video, large language model, DoS audio data transmission
2025

Where Thought Never Lands was filmed on a seabed 20 meters below the surface, where ocean currents etch folds into the sand, forming an alien landscape reminiscent of a desert or the surface of Mars. The camera moves slowly, as if in search of something.

In prehistoric times, humans had language but no writing. The subtitles in this video are generated by the large language model Ollama, which emits them as sound frequencies via DoS (data over sound). Through watching the footage, Ollama begins a stream of self-directed questions—an endless internal monologue that drifts through the unfamiliar seabed, enacting a form of digital thought in motion.

Ollama runs on a closed-system computer, disconnected from any network. Without external input, it resembles a thinker trapped within its own mind. Each time the video loops, Ollama rereads the sentences it previously generated, reflects anew, and continues producing the next round of subtitles. Operating within the constraints of a standalone system, it also tests the boundaries of digital thought. Faced with the 20-meter-deep seafloor, a future two thousand years ahead, and the flow of cinematic time, is the language model’s thinking truly confined—or is it stirring toward an as-yet-unknown edge of consciousness?

Divers: Te-Mao Lee, Jian-Han Lin, Pei-Chin Hsin

<<