Three notes were wrong.
But I no longer had a record of where they were.

Before I could make the practice CD I wanted to finish over the holidays, I needed to locate those notes.

The piece was Handel’s “Hallelujah” chorus. I had entered the score I had into music software to make a practice recording. I had also lowered the whole piece by a whole tone to make it easier to sing.

I knew there were three wrong notes in the score I was using, but the record identifying their locations was no longer at hand. This was meant to be a model for practice, so I wanted to correct the wrong notes before putting it on a CD.

That was why I turned to Kimiko-chan—my nickname for ChatGPT.

I wasn’t asking her to compose something new. I wanted help locating the notes that needed attention in the music data I had prepared.

Once she had helped identify three notes, I changed them in the data. These were pitch changes to individual notes, separate from lowering the whole piece to make it easier to sing.

During the exchange, I added a clarification:

“That’s before the changes.”
“I’ll send the revised one.”

I sent the revised MIDI file—the data used for playback—and explained that it reflected my changes to those three notes.

Making changes on my side is not the same as the other side looking at the revised material. To continue checking, I needed to provide the updated data.

Kimiko-chan helped identify the notes to change. I made the changes and sent the file again. Changing three notes did not establish that the entire piece had been checked, but it was one step toward the practice CD.

Even in a conversation about music, keeping the before-and-after versions straight felt familiar from exchanging data at work.

Before singing, first check that we are looking at the same version. Along with the notes, the conversation needs to stay in step too.

WRITTEN BYYoshio Taki

IBM i / AS/400 System Engineer