The vow

I, j10t, being of sound mind and body, do hereby solemly vow that I shall never use LLM output in my personal writingsi.e. the blogs on this site or in any other form of direct, human-to-human, written contactI need to expand this with a proper paragraph on what this means I think. For me there is a distinction between 'communication' that forms part of business workflow wankery[is there a biz equievelant of the economics 'Econ'?], and genuine attempts to achieved shared understanding via the written word.

Why sweareth thou?

I am not an AI luddite.I pay for the 5x Claude plan (and worry how I will bring myself to ever cancel it). I'm, somewhat concerningly, known as the Claude guy at work. Vibe engineering unlocked me actually building things outside of work! I use claude for everything. Etc etc you get it but I do feel that presenting LLM 'thoughts' as if they were your own is both morally and intellectually repugnant. As a reader I read a human's stuff because I want the human insight, opinions and views. If I wanted LLM transcripts I would prompt them myself.

More controversially perhaps, I think it is almost as bad to let an LLM edit your writing. Spellchecking feels like it is probably okay (but we already have tools for that). Grammar almost invariably comes with re-arranging or changing words, and at that point your authorial voice and intent is lost.IMO, of course.

"Oh, but I just generate a template to get me started and then I edit it" someone says. Nope. Nu-uh. No way. You are still having your thoughts guardrailed by LLM defaults. If you really must, why not just look at another humans output you like and do the old fashioned thing of making those headers yourself?

Exceptions and fuzzy linesAll good chivalric vows came with a list of caveats, right?

While I like the dramatics of a vow, I am not a puritan. There are several sensible exceptions I will allow:

• Generated posts for testing the site - these shouldn't ever end up being seen by end-consumers, and ipso facto has been de jure for testing for eons. LLM output here is fancy ipso that more intentionally tests things.
• Specific LLM instructions to add something specific because I am lazy - e.g. I want to add a link to a specific commit I'm writing aboutsuch as the one that introduces these side notes that I am abusing!. I could go to github myself, browse the tree and copy and paste it. Or I could ask Claude to do it for me. In either case I am in control and directing what goes where, and thus I have no issues with it.
• READMEs- I am genuinely unsure what direction to go on these. A part of me feels like we should try and maintain READMEs as being for human readers, and let the bot-to-bot chat happen in AGENTS.md or equivalent agent specific docs. But that doesn't seem to be the web default - most repos have vibed out READMEs and agents themselves have a strong training-set bias towards looking [and writing] there. So far I have let Claude mangle my README files, and while I've not been happy with it I've also not summoned the energy to re-write them from scratch sooooo....

So what?

Well. Nothing really. This is just a public statement of something I had been thinking about for a while. If I break my own rules, hopefully I'll disclose it (and/or update this post!). But I hope I don't.