Too AI; Didn't Read
Growing frustration with AI-generated “slop” writing is prompting calls for new norms around when and how language models should be used, especially in human-to-human communication. Many argue that long, unedited AI outputs disrespect readers’ time and erode the old expectation that writers do more work than their audience, while others counter that AI can meaningfully help poor or non-native writers express their ideas and that human-written text has always included plenty of low-effort filler. The exchange highlights emerging fault lines over authenticity, effort, and gatekeeping as AI-assisted prose becomes ubiquitous in workplaces and on the web.
Perception of AI “slop” writing
- Many commenters express strong disgust at long, obviously LLM‑generated prose, calling it “slop” that disrespects the reader’s time.
- Common complaints: verbosity, cliché phrasing, overcomplicated explanations, and a homogeneous “Claudish” voice that feels uncanny.
- Some say AI writing is now more annoying than the existence of AI itself or even than enshittified ad-heavy sites.
Use of AI vs. human effort
- Several distinguish between using AI as a private tool (for thinking, translation, or grammar checking) and sharing raw AI output as if it were one’s own.
- A recurrent theme: in human-to-human communication, the sender should do more work than they impose on the reader; dumping unedited LLM output flips this burden.
- Others argue that writing has always involved reuse (memes, ghostwriters), so focusing on “effort” is misguided; what matters is clarity and whether the sender stands behind the content.
Did they even read it?
- Some insist that if someone sends unedited slop, they will assume the sender didn’t read or understand it, unless explicitly stated otherwise.
- Others push back that you can’t know whether they read it, and that dismissing content solely because AI helped write it is unfair.
Access, quality, and ESL considerations
- A number of commenters defend AI assistance for poor writers and non-native English speakers, arguing that dismissing AI-polished text can be exclusionary.
- Others counter that people who rely heavily on AI often can’t recognize its clichés and errors, so their oversight is weak.
AI in code and workplace communication
- Experiences vary: some prefer AI-generated commit messages or summaries over cryptic human ones; others ignore verbose AI PR summaries and read the code directly.
- One practical test suggested: ask an author “why this change vs. X?” to detect whether they actually understand AI-written code.
Detectors, style, and future norms
- Opinions on AI detectors (e.g., Pangram) are split between “useless” and “now fairly precise when they flag something.”
- Several hope models will be trained for brevity and simpler style; some already constrain models via “concise” settings or skills.
- There’s debate on whether complaining about slop will meaningfully slow its normalization or just become another form of noise.