AI Code Policy

The jj-vcs project has a sensible policy that I agree with here: https://github.com/jj-vcs/jj/blob/main/AI_POLICY.md

Abstract

I will readily embrace code contributions developed with modern tools like Copilot, Claude, etc. if they are demonstrably useful. This is a great platform to use as a coding sandbox within a Eurorack/Buchla/Serge modular system, and it is increasingly easy to spin up your own custom algorithms. However, there are just a few things to be said for contributions to the official Phazerville Suite:

• I would prefer no attribution to bots as co-authors.
  • You, the human operator, should take responsibility for the code you have conjured into existence. You own every line.
  • If you haven't sanitized the code (see below), bot attribution is still valid for flagging the slop. Someone else (maybe me) can still use it as a working prototype/example.
• Code comments should be written by humans, for humans. Do not include LLM-generated comments verbatim!
  • Think critically; revise and include only what is necessary for future developers reading the code to understand your intent where it is not already apparent. Especially avoid references to things specific to the context of a conversation you had with your imaginary friend.
• Please squash your work into logical change-sets and keep commit messages brief.
  • A nice one-line summary for the title is usually enough. A few extra bullet points is okay. Thorough documentation belongs in Markdown files or code comments.

Thank you! Keep making cool stuff!

If you're lazy, but you've got something cool that works, I might be willing to sanitize it myself for inclusion, as I do with everything... shoot your shot with a PR. ;)

Sanitation

Anything generated by an LLM should be considered unsanitized by default, even modifications to existing code - it is a hallucination with fuzzy edges. We're trying to clean up the slop. The goal is to have code with 100% verified architecture.

Your task, if you choose to accept it, is to comb thru the generated code and make it your own, so you can justify putting your name on it.

Step 1: Delete All Comments

You can read them to gain understanding, but then you must isolate the actual code, and rewrite your own explanations. Along the way, you may uncover fundamental problems that need fixing. Afterward, you'll be the actual author of all the comments.

Step 2: Sanitize

Try to simplify. LLMs like to grow forests that need lots of trimming.

Step 1 should lead you in the right direction toward verifying the structures. Don't be afraid to fix or completely rewrite things - leave no code untouched! After this, you can rightly claim authorship of all the code.

Step 3: Profit!

Test and verify it, then put your name on it, pal! You've successfully sanitized a piece of hallucinated software! Even if it's a partial effort, perhaps you can grade the Sanitization Level at 50%, or 75%... a graded mark of approval. Human Verified Code.

Expert Mode

Use LLM code for examples, but write everything by hand. Especially great challenge for learning first principles!