After some testing, I discovered that my software has a significant issue. Despite many efforts by Claude to fix, the ‘chords’ command in Finish That Song is still calculating the notes incorrectly – displaying non-diatonic notes as though they were diatonic. You would think that this would be a minor thing to correct but turns out not to be. I’ve actually written similar software myself. The problem is not that hard. It’s just that Claude seems to be caught in some death loop of evolving increasing complexity without ever actually addressing the underlying logical flaw.
Maybe the problem would be easier to fix if I wasn’t constantly reaching enquiry limits and having to restart conversations? Maybe if I could actually attach the whole script for Claude to evaluate and debug? But this is not possible. The script is now too long to be included as an attachment. I guess I could try switching from the free to the pro tier, but why do this if I’m only really aiming to test the system’s potential rather than to engage in anything resembling on-going professional coding activity?
I’m convinced now that it would be better to adopt a less holistic approach – to dispense with ‘vibe’ coding and to gradually build up classes, methods, etc. in a more deliberate, incremental way in which each new addition can be appropriately tested and then slotted into place. In other words, better to conceive the overall system myself and then have Claude assist with aspects of the research, writing and debugging.
It also occurs to me that in terms of my initial goal, I would have much better chance of finishing my songs if I simply made the effort to become more musically literate and analytical myself. I already know the scales and chords. I really simply need to make the relatively small effort to pause and deliberately consider complementary sequence and modulation possibilities. Currently I employ an intuitive method, which often works but can also fail – because I’m tired, lose interest, whatever. Maybe in those moments I should just write the chords down and then return to them later with the explicit goal of examining how they could form the basis of an overall song. Do I really want to be communicating with a command line application in the midst of piano playing and receiving generic advice? Wouldn’t it be more fun and productive for me to allow a mix of intuition and gradually increasing musical literacy?
So the age old question returns – the question that Plato raises in relation to the technical supplement of writing: does writing (the technical aid) augment our internal resources of memory (thinking) or does it diminish them? There is potential, clearly enough, for it to do both and the balance between atrophied dependence and reflective agency is constantly renegotiated in different contexts. But while I play at the keyboard, it seems a better orchestration of my various techno-human parts to keep the process as simple as possible rather than to introduce additional complexity.