AI Stance

Inspired by Armin Ronacher's AI Transparency Statement

I use coding agents every day, but never use LLMs for writing or anything that's meant to be read by humans. I occasionally make a throwaway picture (like an avatar).

I do use them to fix idiotic ESL mistakes I sometimes make. I use … and — via the Compose key in Linux since the 2010s. I have a background in traditional publishing and believe that three dots (...) are not an ellipsis (…) and never will be.

Ground rules

Disclose for humans

Every human deserves to know the source of your information, and has the right to ignore an LLM answer without reading it. It's a good rule even for non-AI applications; I am not reading your "russia today" quote, mate.

So, an LLM response should be marked with the same effort as "trigger warnings" or at least use some quoting standard like APA.

I make an exception for code, because the clanker that did some menial task doesn't deserve a "Co-Authored" byline. Also, no one reads your code anyway (sad, but true).

Accepting full responsibility

An agent/model didn't fuck up. You did.

Never give up control

Don't trust them fully; always have a deterministic pipeline to check their work.

More personally

I dislike the misleading "AI" term and people who are really into it.

Although, I believe any technology can produce obnoxious zealots. Just look at crypto.

Maybe people should look into who raised all of those assholes, instead of screaming at a chain of hashes or matrix multipliers.

I accept LLMs as a genie that can't be put back into the bottle. Holy wars about LLMs are boring.

I do believe that companies behind so-called "SOTA models" are not to be trusted and LLMs should be self-hosted and sovereign.

At the same time, the whole "they drink all our water" trope is a good way to detect virtue-signalling idiots. Most of them can't distinguish an artesian well from an Abyssinian one and have zero interest in the state of local aquifers.

Projectfile

I use the "org.projectfile.ai" extension in my projectfile schema, which generates an "AI POLICY.md" document in all my projects.