I know I could have worked more with ChatGPT to sculpt the output prior to realizing it as sound. I did do a little refining: mainly setting different soprano lines for ChatGPT to use and restricting the scale-pitch choices. Specifying much more was contrary to my original intention -- to feature the AI algorithm itself. I'm also aware of the huge electrical energy toll that many AI operations take. Even though this project represents a ridiculously tiny amount of that energy, it's kind of a ridiculous use of AI, tiny or otherwise. Plus I didn't want to get trapped in the 'tweak the output forever' rabbit-hole. I was more interested in synthesizing the chorales.
The last one (4th one) was very curious, only writing out the same three chords over and over. I think I tried to over-specify the soprano line on that one. I do like it, though!
Piece website: