Developing a Feature Due to ChatGPT’s False Information

Adrian Holovaty shares an unusual experience with Soundslice, a music notation software. The platform’s sheet music scanner, which is designed to digitize images into music, was receiving unexpected uploads in the form of ASCII tablature screenshots. Initially confused, Holovaty discovered that ChatGPT was incorrectly advising users to visit Soundslice, sign up, and upload ASCII tab for audio playback—something that the platform didn’t support. Faced with a growing number of users misinformed by the AI, the team decided to create a feature that allowed for ASCII tab import. This led to the development of a bespoke tool, even though the feature was never on their roadmap. Holovaty reflects on this strange situation, questioning whether it’s a good idea to build features based on misinformation and how companies might need to react to AI-generated falsehoods.