I asked a wide variety of chat bots to write code for Opusmodus.  All of them produced results with mistakes, and all of them repeated mistakes after being corrected.  The free version of chatGPT 3.5 (chat.openai.com) was the best of them all.  It still repeats mistakes sometimes when asked to modify previous code or generate new code, but overall I'm happy with the results.  I still struggle to create original works from scratch with Opusmodus, and this gives me lots of working examples I can generate by describing what I want to make.

chatgpt bach corale3.opmo

funny 😄 but of course it is far far away from bach or a choral harmonization

 

I think there are various difficulties, on the one hand whether chatGPT finds enough information on the net about the harmonization of chorales and probably far too little about OPUSMODUS and LISP.

chatGPT doesn't really manage to program usefully in LISP (e.g. I tested sorting algorithms), but chatGPT explained to me extensively how the resulting code works, but it was always full of errorr/bugs. never worked, not fixable either. there is simply still a large amount of data missing for the system. 

 

But of course I'm not a specialist, I just tried it (chatGPT) out in an amateurish way

Perhaps it's better at coding python.  I used chatGPT to help me write machine learning code for analyzing MIDI files to generate new music.  It works, and it has a GUI interface, but it will take a lot of experimentation and tweaking to get something decent.

The ChatGPT 4 is coding lisp quite well.

okay, then i will try again 😁

Opusmodus Chat GPT has already suggested me two functions, that are obsolete: gen-interval and import-scala-tuning.

 

HenryT

The instruction in Opusmodus GPT quite clearly defined. Try with the prompt to force the gpt to search the knowledge documents first.

So far, my experiences with ChatGPT have been more satisfactory with basic LISP.  Often creating working algorithms of reasonable complexity.  With the Opusmodus GPT, however,  I have yet to get a response that is without programming errors and incomplete results.  Hopefully, people will begin to report any experiences they are having in this regard so we can all learn about how the program might figure into workflows with good results.

OpenAI must improve the handling of provided knowledge documents, as GPT often omits them in responses.