Frustrated by AI’s non-deterministic edits to her sheet music PDFs, a user leveraged ChatGPT to write a deterministic Python tool that reliably removes background color without risking content changes.
- Direct AI edits risk altering original content unpredictably
- ChatGPT can generate deterministic code for precise file handling
- Python libraries enable tailored, batch image processing tasks
What happened
A user’s wife needed to convert yellow sheet music booklets into larger, readable PDF versions without the yellow background to reduce printing ink consumption and improve playback compatibility. Initial attempts to remove the background with Photoshop were cumbersome and inconsistent, with adjustments needed for each page.
The user experimented with having ChatGPT directly edit the PDFs, but the AI’s non-deterministic nature led to low-resolution outputs and subtle, possibly erroneous changes in musical notation, raising concerns about practicing incorrect music. To solve this, the user decided to ask ChatGPT to write a Python program that would handle background removal in a fully controlled, deterministic way.
Why it matters
AI models like ChatGPT produce different results for the same input, which makes them unreliable for tasks requiring exact fidelity, especially in sensitive contexts like preserving musical notation. Relying on AI to modify files directly can introduce errors unintentionally.
By using AI as a tool to generate deterministic scripts, users gain precision and confidence in editing. This approach leverages the strengths of AI – rapid code generation and problem-solving – while maintaining control through classical programming, opening pathways for customized, trustworthy document and image processing.
What to watch next
Similar workflows could become more common, where AI assists by writing specialized software to handle tasks that require strict accuracy and repeatability. Expect growing interest in AI-augmented coding that blends creativity with deterministic execution.
Developers and users should explore the balance between AI’s generative flexibility and the need for predictable outcomes in file manipulation. Advances in AI interfaces that better support deterministic programming or guided code generation could further reduce reliance on trial-and-error direct file edits.