Not Me

Lately I have been hung up on a linguistic difficulty; actually, a phrase: “I, but not me”. As more of my work is done[1] through LLMs, the language of agency and ownership is failing me. Historically, whether I’d done something was a fairly binary concept. I either had - or had not - acted.

Things are more fractional now. I carefully designed each layer of architecture and have a solid conception of what’s happening (Zahir) but wrote little of it myself. Or, I clearly specified what I wanted and wished in the direction of the LLM with a few rounds of “be better” and “be secure” (to.rho.ie). Or, I am in fairly blank incomprehension of basically all of the project [2].

How do I attribute my work?

• I constrained the space of output probabilities such that the text-generator captured novel ideas in functional programming and system design.
• I sent a very good list of demands to an unknowable third-party that mimics a human (though: an odd, amoral LinkedIn sort).
• I cast wishes at a mechanical genie, and remain unsure which were granted.
• The randomly-generated text on my laptop is better than most.

I’m unsure. There’s no simple terminology for “I, circuitously, directed this thing that is some amount of good to exist with this amount of clarity and that amount of noise”.

I found the legal doctrine of “Qui facit per alium facit per se”: “He who acts through another does the act himself”. It’s a beautiful idea: agency is transitive, we keep responsibility not just for our actions, but how we have others act in our place. Glibly, this isn’t common practice: if anything, we’ve migrated maximally far from that ideal.

Put Me Through to a Human!

I find myself having fractional conversations. I (myself, me) ask questions and receive:

You’re absolutely right. Your plan is not just right EMDASH it’s rigorous. Everything you need is in place. Our changes will be safe to merge to production.

Do they agree with the course of action? Certainly possible. They’re probably at least aware of it. Or: Claude is trained on glazers and lickspittles, and fills a sentence with praise like an upstanding person would with “ums” and other disfluencies. It’s hard to tell. They saw my message? It’s not impossible: it was certainly seen - for some extension of that word: a syntactic English response implies as much. Of one thing I can be sure: the changes would blow things to bits. People aren’t to be believed when proclaiming safety, much less their impersonators.

So, things devolve. Writers convey ideas: sometimes to enlighten, sometimes to deceive. The LLM fanatic instead types something hilariously half-assed into a machine that dresses up their cloth-sack sentence into something splendid. The result is detailed, professional, formal, and devoid of meaning. It’s mimicry; obtuse walls of text camouflage the indifferent and work-shy just as “X is a Land of Contrasts” and “Webster’s Dictionary Defines…” has for generations hid students’ patchy knowledge from their less vigilant examiners. Little intent is conveyed and no thought is signified. Who knows what they actually typed into their LLMs. My guesses, though, horrify me.

In Summary

I do not have the terminology for people nodding off while supervising dynamite factories [3]. Or for those half comprehending conversations and quarter participating in them. It seems like necessary language, though, and something will turn up. It’ll probably come from 4chan [4]

Footnotes

[1] These are not successful

[2] done being equally tricky, having become a floating signifier that’s drifted from the old norms of “existing”, “functioning”, and “comprehended by the doer”.

[3] the wording would be very clear if they weren’t typically under orders to do so.

[4] Yes I do see the em-dashes. I worry this too may be Pod-Person literature.