scientific-prose

Scientific prose

Banned patterns

• Em dashes (—) and en dashes (–): em dashes are clear signs of AI writing, so remove them and replace them with more natural alternatives, like periods or commas. If not possible, restructure the sentence. This is a hard constraint.
• Throat-clearing openers: "It is important to note that...", "It's worth mentioning..."
• Unearned transitions: "Moreover", "Furthermore", "Additionally" used as connective filler rather than because the next sentence needs that specific logical relation
• Preview sentences: "In this section, we will discuss...". State the content directly instead.
• Inflated vocabulary where a plain word means the same thing: "delve", "leverage", "robust", "multifaceted", "testament", "tapestry", "comprehensive" (as a filler adjective), "utilize" (use "use"), "load-bearing", "align with", "crucial", "emphasizing", "highlight" (verb). Additional AI-frequent terms to eliminate: "actually", "interplay", "intricacies", "landscape", "fostering", and "garner".
• Manufactured rule-of-three lists and uniform sentence length: vary structure the way a written derivation or argument naturally does
• Hedge words used to avoid a claim rather than to state genuine uncertainty: "may", "could potentially", "generally speaking": either state the result or state precisely what is uncertain and why
• Generic conclusions: "This shows the importance of X" — end on the specific finding, not a platitude
• Unnecessary comparisons: "We do X rather than Y", "It is X, not Y", "What it does, what it does not". If there is no need for a direct comparison, avoid it: just state plainly the central idea.
• Undue emphasis on significance and legacy: avoid framing standard methods or incremental results as a "pivotal moment" or "vital role". Avoid saying a development "marks a shift" or is "setting the stage for".
• Superficial analyses with "-ing" endings: avoid tacking present participle phrases onto sentences to add fake depth. An example of this is writing "achieving SOTA", "underscoring its efficiency".
• Promotional and advertisement-like language: avoid "groundbreaking", "renowned", or "boasts a" when describing architectures or experimental results.
• Vague attributions and weasel words: do not use "Experts argue", "Observers note", or "Industry reports". Ensure you attribute opinions or prior findings with specific sources.
• Outline-like boilerplate sections: avoid formulaic "Challenges and Future Prospects" structural framing in the discussion.
• Copula avoidance: do not substitute elaborate constructions like "serves as", "represents a", or "offers a" for simple verbs like "is" or "are".
• Negative parallelisms and tailing negations: avoid overused "Not only... but..." constructions. Do not use tacked-on fragments like "no guessing" in place of complete clauses.
• Elegant variation (synonym cycling): do not unnecessarily swap technical terms (e.g., cycling randomly through "model", "network", and "framework") just to avoid repetition.
• False ranges: avoid "from X to Y" constructions where X and Y are not placed on a meaningful scale.
• Signposting and announcements: omit meta-commentary like "Let's break this down" or "Here's what you need to know".
• Diff-anchored writing: describe the methodology or architecture as it currently is. Avoid narrating the process of changing it from a previous version unless explicitly writing a changelog.
• Aphorism formulas: avoid reducing rigorous technical claims into pseudo-profound formulas like "X is the Y of Z".