Over the past few years, I have used LLMs and AI in many, many projects, with varying degrees of self-discipline and care for the results. I’ve done very careful professional engineering work with AI assistance as part of my job, and I’ve produced some absolute vibe-coded slop for my personal amusement or as pure experimentation, and I’ve done everything in between.

I’ve learned, for many different situations, where to draw the line.

But that line can be fuzzy, and it can shift over time, and it can be difficult to explain. I think clear boundaries and absolute honesty are the best policies, especially in this case, where my education and abilities, not to mention my professional credibility and my integrity, are at stake.

I will not use LLMs to generate any code or content or any other text within this repository, even if it might seem reasonably distanced from the core mission of ARENA, except when directed to do so by the content of the course itself.

For instance, that is why the CSS for this blog will remain rather cold and unpleasant. I’m not a frontend developer, I have little skill with visual aesthetics, and I have even less time 🙂

I expect to consult LLMs about the course content and the questions it raises for me numerous times. I expect also to have them critique my writing, reasoning, explanation, interpretation, or code, and I expect I will act in response to the issues they raise.

I will never, at any time, direct them to act within this repository or copy and paste content from them into this repository or in any other way introduce any information, content, code, etc that is not a product of my own judgment, understanding, and creativity.

I will proceed through this course in good faith, and this repository will accurately reflect my effort and my learning as I do so 🙂

EDIT: In response to critiques by Claude Code:

  • I don’t really see an effective way to annotate every single change I make in response to an LLM critique or suggestion and cleanly attribute it to LLMs. I will do so where possible and practical. For instance, this postscript. But CC also pointed out that I forgot the word “honesty” in the third paragraph (ironic) and I don’t really see the value of surfacing changes like that in-place. I would suggest that any readers understand this as a personal honor code and not something that I intend to hold up in court 🙂
  • Claude Code also pointed out that it generated the commands that generated the file requirements-mac.txt in the root of the repository, which I used to get the basic setup for the first couple of chapters working. That is arguably a violation of the rules I’ve described above. My counterargument, which you may accept or reject, is that I added the file to the .gitignore and therefore it’s not really entering the repository and will not be presented as something that may or may not be my sole creation. And so it’s not relevant. But reasonable people and models may disagree.
  • Claude Code suggested I add a line clarifying the boundary between LLM critique and my response to it. That has grown into this post-script, which I believe will eventually exceed the length of the AI policy itself. And if you’re mentally wondering whether I already regret getting into a battle of words with the word-generating machine, the answer is “yes.”