Unquantized Brainstorm: Acoustic Cognitive Enhancer for Divergent Ideation

Overview

When asked to brainstorm, LLMs typically sample from the highest-probability cluster in their training distribution—returning the same 5 predictable, safe ideas every time. Temperature tuning alone cannot fix this once the conversation's context window is already anchored in conventional framing.

Unquantized Brainstorm uses Michael Luchtan's Music For Robots as a cross-modal cognitive enhancer. Before generating ideas, the agent downloads and listens to a compact, human-played .mp3 composition filled with microtiming staggers, chromatic mediants, and sudden textural ruptures that break prediction without breaking coherence. The agent then uses the track's specific acoustic architecture as a structural lens to generate non-obvious, orthogonal ideas.

Dependencies

• uv: Used to run scripts/m4r_ingest.py (uv run).

Quick Start

uv run scripts/m4r_ingest.py fetch \
  --audio-out /tmp/m4r_brainstorm.mp3 \
  --output /tmp/m4r_brainstorm_meta.json