AI Slop — One Paper, Two Rebuttals
Dan PetersonSynopsis — AI-drafted from Dan's notes
Three works on the same question: is AI-generated content worth anything? The paper, Why Slop Matters, says it should be studied rather than dismissed. Near-zero production cost means content no longer has to be good to be worth making, so slop supplies niche demand nothing else could, and like kitsch it can carry real meaning for the people who consume it. Meaning, on this view, is made on reception. The authors warn that critics who treat slop as uniquely malignant are repeating Clement Greenberg’s 1939 dismissal of work that was later canonized.
Better Offline’s episode answers it directly. AI output may have social uses, the hosts argue, but it is not art: art needs the friction of learning a craft, mastery of the rules before breaking them, and a human who meant something. Prompting makes you a patron, not an artist, and much of the talk of democratization is cover for monetization.
David Gerard comes at it from the market. AI content is pushed by companies rather than wanted by people, and it is rejected because it is low in information: statistically average words that skip the thinking-out-loud writing is for. He reads the backlash on platforms and marketplaces as the evidence.
Where I land
I thought the Better Offline episode was a very thought-provoking argument. In essence, art requires experience and viewpoint. I don’t know if LLMs, even if they are seeded with the world’s knowledge, or agents programmed with a persistent memory or viewpoint, count as “experience”: actually lived in history, in the world, with real consequences and social connection. Part of art is not only the creation but also the reception by a fellow human being who can connect to the material on some level.
On friction, like any skill, the painful part signals growth.
A counter to David Gerard, who in this video said that LLMs have no use as they are simply word-guessing machines, and as such AI-generated content is a marker of low quality: I would argue that there are insights that are valuable, like the sessions behind this garden. Purely AI-generated work (with no context, direction, feedback or craft) does present low-quality work. But AI-assisted work, with rounds of iteration and human viewpoints and judgement, could be demonstrably better output. It’s almost like comparing hand-written notes with notes typed in a word processor with spell and grammar check. Same writer, but I would argue the one armed with the latter will be more productive and produce better output than the former.
Connections
Links
- Cody Kommers, Eamon Duede, Julia Gordon, Ari Holtzman, Tess McNulty, Spencer Stewart, Lindsay Thomas, Richard Jean So and Hoyt Long, “Why Slop Matters” (arXiv, 2025-12-23) — https://arxiv.org/abs/2601.06060
- Better Offline, “LLMs Will Never Make Art With Caleb Wilson and Arif Hasan” (2026-09-03) — https://www.youtube.com/watch?v=gKuIYnRKRTU
- The Tech Report, “Nobody wants to buy something AI generated | David Gerard” (2026-09-02) — https://www.youtube.com/watch?v=vkIn0Xh0Ld8