Has AI tells
Cheap, infinite AI-generated text is reshaping online writing, prompting efforts to detect “AI tells” and filter out formulaic content. Commenters debate whether provenance matters more than clarity: some prioritize human effort, authenticity, and writing-as-thinking, while others care mainly about usefulness, especially for technical documentation or translation. Several simple rule-based tools and heuristics for spotting LLM-style prose are mentioned, alongside skepticism about their accuracy and concern that future models will be harder to distinguish from human writers.
Nature of the “AI tells” tool
- No public link; described as an in-house, rules-based filter used to label/weed out AI-ish articles in a curated feed.
- Implemented as ~100+ hand-written rules comparing pre- vs post-LLM era writing (around 2023), with fixed thresholds.
- Used because running a heavier detector (Pangram) on every item is too expensive at the scale of ~23k+ articles.
- Purpose: keep low-quality/marketing/AI-slop out, support tagging/ranking of technologies and topics.
Other tools and detection challenges
- References to similar or related tools: “is-it-slop”, an LLM cliché highlighter, and research like SlopShape for identifying both human vs bot and specific models.
- Some detectors misfire on formatting (e.g., markdown lists flagged as repetition).
- Recognition that this is a cat-and-mouse game: detectors vs tools that make AI text look more human or personalized.
Value of human vs AI writing
- Many commenters insist authorship matters: human-written text has intrinsic value, conveys lived experience, effort, and genuine communication.
- Others argue we should judge only by usefulness and clarity, especially for technical documentation.
- Several see effort as a key filter: they refuse to spend more effort reading than the writer spent producing.
- Some frame it as an “attention war”: infinite cheap AI output forces readers to guard attention with stricter filters, including “no/less AI” rules.
Quality, style, and “AI voice”
- Common complaints: verbosity, over-explanation, repeated stock patterns, lack of real audience model (“theory of mind”).
- Some note that AI can already correct human mistakes and improve clarity, and predict future models will be harder to distinguish.
- Others find minor human typos and awkwardness reassuring signals of authenticity.
- Suggestion to “desloppify” even human writing via LLMs, while acknowledging AI has “soured” certain once-neutral phrasings.
Ethics, culture, and crowding-out concerns
- Fears that near-infinite AI “slop” will crowd out human expression in writing, research, and art, and degrade search and feeds.
- Counterpoint: attention and existing filters already handle abundance; AI-generated content will simply compete on quality.
- Some highlight “say–do” gaps: people claim to hate AI writing, but detectors show a lot of it being upvoted and promoted.