Slop (originally the watery food given to pigs) is low-quality content mass-produced with AI: absurd images circulating on social media to harvest engagement, spurious books in online stores, filler articles written for search rankings, generic replies where a person was expected. The term was in use from around 2022 on forums such as 4chan and Hacker News; in May 2024 the programmer Simon Willison popularized it with a comparison that stuck: just as spam named unwanted email, slop names generated content that nobody asked for (Willison, 2024). Merriam-Webster made it its 2025 word of the year and defined it as low-quality digital content produced, usually in quantity, by means of AI.
The concept is more economic and aesthetic than technical: the problem is not that a machine wrote a text, but that its production cost has fallen to almost zero while the cost of reading it remains human. That is why slop is an evaluative term as well as a descriptive one, and why it also applies to human texts imitating the generic prose of chatbots. It does have a technical consequence: if the web fills up with generated text, future models will be trained on it (corpus), and Shumailov et al. (2024) showed that recursive training on generated data can degrade models to the point of "collapse," as the tails of the distribution disappear.
Defenders of these tools reply that slop is a judgment about use, not about the technique. The question turns reflexive with art: the p(doom) song itself was generated with Udio, a music model, over lyrics written by people with help from Claude (osmarks, n.d.). Is it slop? For philosophy of language, the term is a good case study: if meaning depends on the speaker's communicative intention (Grice, 1957), and models lack one (Bender et al., 2021), where does a reader of slop get sense from, and whose sense is it (form vs meaning)?