I used ChatGPT to build a free PDF editor because I didn't trust it to change my files - it's glorious
Original reporting by ZDNet
When his wife, Denise, joined the church choir, ZDNET's David Gewirtz found himself tackling a surprisingly tricky problem: transforming yellow-tinted sheet music into usable, eye-friendly PDFs. Denise needed to print enlarged versions for easier reading without glasses and feed them into a music playback app, but the original yellow background meant significant ink waste and potential software glitches that could misinterpret the notes. Initial attempts to manually remove the color in Photoshop proved too "fiddly," requiring tedious, inconsistent adjustments for each page.
The AI's true role Gewirtz then tried a more advanced approach, asking ChatGPT to directly process the PDF files. While the AI successfully removed the yellow, a critical issue arose: ChatGPT, like many generative AIs, is "non-deterministic." This meant each attempt yielded subtly different results, reflecting a probabilistic rather than exact output. For something as precise as sheet music, where even minor alterations could lead to incorrect practice, this lack of reliability was an unacceptable risk. The solution, he realized, lay not in having AI directly perform the task, but in having it *build a reliable tool*. Gewirtz prompted ChatGPT to write a Python script that would deterministically remove background colors from multi-page PDFs. The result was a simple, command-line program that flawlessly delivered clean, white-background sheet music, preserving every crucial note. This story illustrates a powerful lesson: AI's most profound contribution isn't always in its direct output, but in its ability to quickly craft the precise, deterministic software we truly need.
The specific challenge of accurately removing background color from scanned sheet music, while meticulously preserving its integrity for both human practice and digital playback, was elegantly resolved not through direct AI processing, but by leveraging ChatGPT as a sophisticated code-generation engine. This practical demonstration underscores a critical distinction in AI utilization: employing a non-deterministic AI to craft a reliable, deterministic Python script. This method successfully delivered a customizable, efficient solution that directly addresses prevalent concerns about AI’s inherent variability in sensitive applications where absolute accuracy is paramount.
A new paradigm for tools
This strategy carries profound implications for the evolving relationship between humans and artificial intelligence. It positions large language models as powerful productivity multipliers, empowering individuals – even those without deep programming expertise – to rapidly develop bespoke tools tailored to highly specific needs. By efficiently generating robust, deterministic code, AI can tackle a vast array of precise, repeatable tasks, significantly reducing traditional development cycles and democratizing access to custom automation. Looking forward, we can envision a future where AI-generated micro-tools become commonplace, transforming personal productivity, streamlining workflows across diverse industries from academic research to small business operations, and fundamentally altering the landscape of software development. This collaborative model, where AI serves as an intelligent co-creator of functional applications, amplifies human ingenuity and accelerates problem-solving, moving beyond simple content generation to enable a new era of highly customized and efficient technical solutions across the board.