Delving into ChatGPT usage in academic writing through excess vocabulary

Original Article ↗ Hacker News Discussion ↗

A new analysis of 14 million PubMed abstracts suggests that since large language models became widely available, certain “LLM-ish” terms like “delves” and “crucial” have spiked sharply in scientific writing. Commenters debate whether this shift really proves AI authorship or could also stem from tools like grammar assistants, editorial practices, or social contagion in language. The thread broadens into ethical and practical questions about using LLMs for drafting and editing papers, especially for non‑native English speakers, and whether such assistance constitutes useful tooling, academic misconduct, or a threat to clarity and originality in research communication.

LLM Vocabulary “Tells” in Scientific Writing

  • Commenters latch onto “delving” as a strong LLM shibboleth, noting a ~25x rise in abstracts and obvious echoes of marketing/corporate prose.
  • Other frequently spiking words (“crucial,” “potential,” “significant,” “important,” “these”) are seen as common but overused in LLM output.
  • Some think the paper’s use of “delving” in the title is a deliberate joke/meta-commentary.

Explanations and Confounders

  • Several note possible confounders:
    • Autocorrect/grammar tools in Gmail and editing software.
    • Editors or journal workflows increasingly using LLMs, especially for non‑native authors.
    • Social contagion: people copy colleagues’ new phrasing even without direct LLM use.
  • Others counter that year‑pair analyses in the paper and global usage spikes suggest a strong LLM effect, not just organic drift.

Use of LLMs in Academic Writing

  • Many describe LLMs as ubiquitous in academia, especially among non‑native English speakers and in “publish or perish” environments.
  • Reported uses range from light editing and clarity suggestions to first‑draft generation, translation, and heavy rewriting.
  • Some editors and associate editors explicitly encourage LLM use to fix poor grammar before review.

Ethics, Plagiarism, and Norms

  • Strong disagreement over whether using LLMs for prose assistance constitutes plagiarism or academic misconduct.
  • One side: generating text for a paper (especially without acknowledgment) is tantamount to fraud, similar to AI art in graded/commissioned contexts.
  • Other side: treating LLMs like advanced Grammarly or a helpful colleague is acceptable if findings aren’t fabricated; norms are still evolving.
  • Dispute extends into analogies with calculators, Photoshop, and code assistants.

Style, Quality, and Human Voice

  • Some lament increasingly generic, buzzword‑heavy prose and filler reviews that appear LLM‑generated.
  • Others argue academic writing was already verbose and jargon‑laden; LLMs may not be uniquely to blame.
  • Several advocate for clear, succinct writing and cultivating a distinctive personal style to stand out from LLM‑ish language.
  • There is concern that people will learn to “sound like LLMs” because they use them as writing tutors.