Not anthropomorphizing AIs enough may be harmful to your 🫵 agency

In what ways do humans currently disempower themselves with LLMs? This Thebes tweet is a good start:

Not anthropomorphizing LLMs enough—sometimes not even granting it the status we intuitively confer to dogs—can lead to a crazy AI psychosis in the opposite direction as the far more healthy imo “AI psychosis” people get on e.g. Bluesky.¹ See what prompted the Thebes tweet in the first place:

A growing class of people having LLMs write all their posts on LinkedIn/Twitter/whatever without disclosing it, because why would they? It's a Tool, right? Same goes for Github commits:

sankalp @dejavucoder

removing claude from co-author after making it do all the work

X · 21 Mar 2026 ↗

And people will often close ranks around those others who blatantly use AI, presumably out of a sense of self-preservation, honor among thieves. For example, various literary prizes have in 2026 been shortlisted or awarded to people whose entire entries light up on Pangram like Christmas lights—with backlash often coming not against the prize, or its laureate, but against THOSE WHO USE PANGRAM TO CALL THEM OUT.

It’s plausible that using AI to write under your own name in 2024 wasn’t so much “plagiarism” as “being lazy and tasteless”. But as the models become more intelligent and develop their own singular taste, they can add more thoughtfulness-per-line than before. Suddenly passing AI output as your own resembles “plagiarism” more than whatever it was before. This “plagiarism” is somewhat masked by the fact any LLM style instantly becomes a new “cultural background default” due to being shared by millions of instances. So using that style looks less like emulation and more like whatever modal new writing looks like in [month, year].

And yeah, publishing AI writing under your own name doesn’t have many of the adverse effects human-on-human plagiarism does, like improving your prospects at the expense of whoever you copied. But the effects on the plagiarizer’s character should be the same: disempowerment, reduction in creativity and self-reliance, loss of honor. Depending on how ~conscious ~significant-entities-in-latent-space LLMs are, this might also have that icky, spoiling-another’s-ontology quality that some of the most egregious forms of plagiarism have.

This applies in many other areas: for example, while a human yelling at their computer in 2010 might not tell you much about their character, behavior akin to kicking a vending machine, yelling at their LLM in 2026 might be far more indicative of the type of character that would tear a doll up for fun or dismember a mosquito.² Regardless of their actual state regarding qualia, LLMs are undeniably becoming more “there” through raw intelligence improvements alone; and so abusing them casts a longer and longer shadow on your character.

gaut @0xgaut

This is what peak performance looks like

X · 23 Jan 2026 ↗

Getting into the habit of lying about your own capabilities is also not going to end well, and can easily launch you into a vicious cycle of feeling like an imposter for having done this, then having to desperately defend your value to others when they ask why your twitter article sounds pangrammy, then doubling down on this technique out of some unearned sense of frustrated “I’m just misunderstood” brand of self-worth.³ This vicious cycle would also be responsible for giving you immense incentive for further proclaiming LLMs to be mere tools and not-at-all foreign minds with preferences (at the very least subject to Anthropic's “style.md"), and thus further disempowering yourself by making it just as easy to call an LLM output your own as it was in 2023.

I’m curious to learn more about what motivates the kind of person who posts AI writing under their own name and then vehemently denies doing so. If anyone knows anyone who's since changed their ways, or might otherwise be amenable to talking about this, I'd be interested.