Ai and Intent

Summary

I think what has emerged for me as the difference when I want and think is ok to use mostly AI based output is:

If I (or you) am communicating to a computer, as in writing programs, scripts, config files, etc. Then usually AI is good enough nowadays.

On the inverse if I am communicating to a person, be it this blog post, be it an email, a document for the management etc, then I need to write that myself.

Gray areas?

The places where this overlaps one should write for the humans (i.e. write the thing yourself). Even if there is a PR that I was AI generated, I should read and understand it as if I wrote it if I expect someone else to read and understand it. If the descriptions are meant for a person, then write them yourself.

I also view certain types of prompts as just smarter versions of some tools. For example a big LLM can spellcheck my writing much better than I can. It can take my crappy vizualization of some data and make it pretty(er). In this cases there are other tools for all of this, I can just get a slightly better result with AI.

Context

I was able to put this into words for myself with the context of no slop grenade and Building Pi With Pi, specifically this part:

Unfortunately, it does not fully work, because when humans first throw their issue through the clanker wringer, their clanker expands scope almost immediately. What was once a very narrow and fact based bug observation, turns into a much expanded surface area full of hypotheses. So at least personally, I increasingly want issue reports to be condensed to what the human actually observed:

I ran this command.
I expected this to happen.
This happened instead.
Here is the exact error or log.

LLM contributions to repos

I am adding text to the following effect to all repos that I maintain in the near future:

Contributions

Bugfixes are welcome but new features will be simissed out of hand if not peviously discussed with the maintainers

LLM Use

You are welcome to use LLMs to generate code. Any PRs texts and comments need to be written without the use of LLM, and the submitter is expected to understand the submitted code. Violation of either will result in a rejected PR.

Footnote: This is part of a much longer ramblings on AI that may or may be published at a later time.