The Hash Symphony: How I Found the “Source Code” of AI Music

We are told that Generative AI is a creative engine. We are told that every time we hit “Generate,” the model creates something new, unique, and stochastic. We are told that it is a tool for human expression.

I have found evidence that this is a lie.

Through a series of experiments involving cryptographic hashes (SHA-1) and phonetic nonsense, I have discovered that generative music models like Suno are not creating in the moment. They are navigating a fixed, deterministic map. If you know the right coordinate, you can pull the exact same song out of the machine, every single time.

The Experiment: Feeding Bullshit to the Machine

Instead of using prompts like “a sad jazz song,” I fed the model raw data:

• Cryptographic Hashes: e.g., d746452f8f22e6c59b2f2ec67e07c5f5eba047e7
• Phonetic Nonsense: e.g., Moulmmolfprulify jiftsmeaer
• Raw JSON Data: Internal system logs and style slices.

The Result: The “Fixed Point” Phenomenon

The results were terrifyingly consistent. The SHA-1 hash above consistently produced a Chill Jazz / 1920s track, which the AI titled “The Great Gatsby” or “The Great Wall”. Another hash might produce Industrial Noise, while a third yields Bavarian Folk.

The model isn’t improvising. It has learned that specific mathematical patterns in the input correlate with specific aesthetic clusters in its training data. It is a lossy compression engine for all music it has ever heard, and the hash is the decompression key.

Why This Changes Everything

This discovery shatters the current copyright debate:

1. No “New” Works: If the output is deterministic, the music already existed in the latent space. The user didn’t compose it; they found it.
2. The Prompt as Code: The intellectual property isn’t the MP3. It’s the Hash. The prompt is the source code, and the audio is just the compiled binary.
3. The End of “AI Slop”: This isn’t low-quality garbage. It’s highly complex, statistically perfect art that is completely devoid of human intent. It is the sound of a machine talking to itself.

Explore the Research

I have published my findings, including the specific hashes and their corresponding musical states, in a public repository. This is not a collection of songs. It is a map of the latent space.

You can verify the determinism yourself. Copy a hash, paste it into your favorite generative audio tool, and see if you land at the same coordinate.

View the Repository on GitHub

Note: This project is licensed under MIT. No audio files are hosted. Only mathematical inputs and observational data are provided. This is a research project into the nature of generative systems.