A Brief Thought on LLMs and Shame

A staff engineer at work posted a dashed-off proof of concept. It was an interesting idea, with just enough functionality complete to point in the right direction. The engineer ended their announcement of the PoC with “it’s got tons of bugs, it’s entirely vibe-coded”.

This is fine and good. Proof of concept efforts are extremely valuable. “Shitty first draft” applies equally well in writing code as in writing prose. PoC || GTFO and all that stuff. As with all rough drafts, putting a PoC out into the world is uncomfortable. You’re intentionally exposing your shitty initial work to early feedback. I think that LLMs serve as a buffer to this emotional vulnerability and discomfort. After all, if it’s not your work, but rather “the machine’s work”, that shame is discharged.

I think this explains something of the adoption of LLMs to do writing by people who, ostensibly, want to be writers. By artists who want to be artists, etc. The shitty rough draft feels bad. Getting feedback on it feels excruciating. Being bad at something you want to be good at is one of the hardest emotional challenges that humans have to overcome in order to do anything of meaning with their lives.

We don’t have a healthy relationship with failure in our society. (Any detailed examination of why this is is out of scope of this post, but the spoiler is a sick blend of individualist culture, social media, and Capitalism). The shame of being bad at something, of maybe putting something bad or even rough around the edges out into the world, can feel overwhelming. LLMs serve to discharge that shame. Any “bugs” in the art are ascribable to the machine, not the “artist”. They give us an ablative layer of blame deflection so that we never have to wrestle with the fear, shame, and self-doubt of being a neophyte artist.

It also bears mentioning that this shame and self-doubt doesn’t really go away as we improve our craft. We get better ways of coping with it (the “shitty first draft” mantra is, itself, a way to help artists move beyond that). Our functional artistic skills increase and we learn how to avoid some of our own artistic failure modes. But artists are always better critics than they are artists. We are all audience members for art before we are creators of it, and because we all take in much more art than we produce, our critical faculties are always stronger than our artistic ones. As Gene Wolfe said: “you never learn how to write a novel. You just learn how to write the novel you’re on.". So the shame and uncertainty is there forever.

This helps explain why even some established authors and artists are tempted to AI. Writing well is hard. Writing honestly is hard. The fear and self-criticism of writing something daring can be overwhelming. If it’s the machine’s words, and not my own, suddenly that shame is lessened and the artist doesn’t have to hate the thing they produced. After all, it’s not really their work.