The revolt of the reader
Growing use of large language models to generate blog posts, documentation and other prose is prompting readers to push back against what many describe as uniform, shallow “slop.” Commenters debate whether AI-generated text is actually easy to detect, how reliable tools like Pangram are for flagging it, and the risks of false positives in settings like education or hiring. Underneath is a broader concern that outsourcing writing erodes both the writer’s thinking process and the reader–writer “social contract,” even as some argue that AI is acceptable for limited support roles such as research, proofreading, or help for non‑native speakers.
Backlash Against LLM-Generated Prose
- Many commenters describe a strong aversion to “LLM slop”: verbose, cliché‑ridden, meandering text that feels overblown relative to the topic.
- Common complaints: same generic “assistant voice” across different authors, formulaic constructions (“it’s not X, it’s Y”), overdramatic tone, and repetition that makes pieces cognitively tiring.
- Several people say they now bail out of any article, blog post, README, or YouTube script that “smells” like LLM output and see this as a time‑saving filter.
Detection and Pangram
- The AI detector Pangram is widely discussed. Some users say it aligns well with their own intuitions and is useful for auditing their own public writing.
- Others report brittleness: small edits flipping human ↔ AI scores, high false positives on clearly human text, sign‑up frictions, and warn it should never be used in education or for accusations.
- There’s demand for cheap or open alternatives and browser extensions that flag or fade likely‑AI text; a few prototypes and wrappers are shared.
- Some object in principle to “outsourcing slop detection” to a model, arguing that “slop is self‑evident.”
Use of LLMs in Writing & Workflows
- Productive uses endorsed by many: research, clarifying ambiguous source material, grammar/spell checks, technical error spotting, summarization, help for non‑native speakers—explicitly without letting the model rewrite prose.
- Others happily use LLMs for drafts, pitch decks, thumbnails, and specs, claiming overall quality and business focus improve, even if style is samey.
- A recurring argument: if you need an LLM to generate the draft, maybe you didn’t have enough to say.
Workplace, Education, and Social Norms
- Multiple reports of coworkers generating bloated specs and communications via LLMs, which collaborators find exhausting and disrespectful; some suggest refusing to engage with AI‑mediated messages.
- Strong support for the idea that writing is thinking: offloading prose to a model weakens the author’s understanding and breaks an implicit social contract that the writer should bear most of the cognitive load.
- Others are indifferent to provenance, caring only about clarity and accuracy, and argue that high‑quality AI text is acceptable if properly disclosed.
Style, Taste, and Ability to “Tell”
- Commenters disagree on how reliably humans can detect AI text. Some claim it’s “blindingly obvious”; others, including linguistics‑aware participants, say informal “AI tells” are unreliable and mimic witch‑hunt dynamics.
- Discussion of classic prose (e.g., Ecclesiastes vs a modern paraphrase) illustrates that preferences for “color” vs “clarity” are highly subjective—and that LLM‑like “bureaucratic fuzz” is precisely what many readers now revolt against.