diff --git a/prompts/LLM_PROSE_TELLS.md b/prompts/LLM_PROSE_TELLS.md index 19898aa..26ce010 100644 --- a/prompts/LLM_PROSE_TELLS.md +++ b/prompts/LLM_PROSE_TELLS.md @@ -22,11 +22,9 @@ reading. Even outside the "not X but Y" pivot, models use em-dashes at far higher rates than human writers. They substitute em-dashes for commas, semicolons, -parentheses, colons, and periods, often multiple times per paragraph. A human -writer might use one or two in a piece for a specific parenthetical effect. -Models scatter them everywhere because the em-dash can stand in for any other -punctuation mark, so they default to it. More than two or three per page is a -signal. +parentheses, colons, and periods. A human writer might use one or two in a +piece. Models scatter them everywhere because the em-dash can stand in for any +other punctuation mark. More than two or three per page is a signal. ### The Colon Elaboration @@ -52,7 +50,7 @@ bother maintaining. Runs of very short sentences at the same cadence. Human writers use a short sentence for emphasis occasionally, but stacking three or four of them in a row -at matching length creates a mechanical regularity that reads as generated. +at matching length creates a mechanical regularity. ### The Two-Clause Compound Sentence @@ -60,7 +58,7 @@ Possibly the most pervasive tell, and easy to miss because each individual instance looks like normal English. The model produces sentence after sentence where an independent clause is followed by a comma, a conjunction ("and," "but," "which," "because"), and a second independent clause of similar length. Every -sentence becomes two balanced halves joined in the middle. +sentence becomes two balanced halves. > "The construction itself is perfectly normal, which is why the frequency is > what gives it away." "They contain zero information, and the actual point @@ -71,8 +69,7 @@ sentence becomes two balanced halves joined in the middle. Human prose has sentences with one clause, sentences with three, sentences that start with a subordinate clause before reaching the main one, sentences that embed their complexity in the middle. When every sentence on the page has that -same two-part structure, the rhythm becomes monotonous in a way that's hard to -pinpoint but easy to feel. +same two-part structure, the rhythm becomes monotonous. ### Uniform Sentences Per Paragraph @@ -80,14 +77,13 @@ Model-generated paragraphs contain between three and five sentences. This count holds steady across a piece. If the first paragraph has four sentences, every subsequent paragraph will too. Human writers are much more varied (a single sentence followed by one that runs eight or nine) because they follow the shape -of an idea, not a template. +of an idea. ### The Dramatic Fragment Sentence fragments used as standalone paragraphs for emphasis, like "Full stop." or "Let that sink in." on their own line. Using one in an essay is a reasonable -stylistic choice, but models drop them in once per section or more, at which -point it becomes a habit. +stylistic choice, but models drop them in once per section or more. ### The Pivot Paragraph @@ -102,14 +98,12 @@ Delete every one of these and the piece reads better. > "This is, of course, a simplification." "There are, to be fair, exceptions." Parenthetical asides inserted to look thoughtful. The qualifier never changes -the argument that follows it. Its purpose is to perform nuance, not to express a -real reservation about what's being said. +the argument that follows it. Its purpose is to perform nuance. ### The Unnecessary Contrast Models append a contrasting clause to statements that don't need one, tacking on -"whereas," "as opposed to," "unlike," or "except that" to draw a comparison the -reader could already infer. +"whereas," "as opposed to," "unlike," or "except that." > "Models write one register above where a human would, whereas human writers > tend to match register to context." @@ -120,8 +114,7 @@ still says everything it needs to, the contrast was filler. ### Unnecessary Elaboration -Models keep going after the sentence has already made its point, tacking on -clarifying phrases, adverbial modifiers, or restatements that add nothing. +Models keep going after the sentence has already made its point. > "A person might lean on one or two of these habits across an entire essay, but > LLM output will use fifteen of them per paragraph, consistently, throughout @@ -129,9 +122,9 @@ clarifying phrases, adverbial modifiers, or restatements that add nothing. This sentence could end at "paragraph." The words after it just repeat what "per paragraph" already means. Models do this because they're optimizing for clarity -at the expense of concision, and because their training rewards thoroughness. -The result is prose that feels padded. If you can cut the last third of a -sentence without losing any meaning, the last third shouldn't be there. +at the expense of concision. The result is prose that feels padded. If you can +cut the last third of a sentence without losing any meaning, the last third +shouldn't be there. ### The Question-Then-Answer @@ -167,16 +160,15 @@ becomes "craft." The tendency holds regardless of topic or audience. "Importantly," "essentially," "fundamentally," "ultimately," "inherently," "particularly," "increasingly." Dropped in to signal that something matters, -which is unnecessary when the writing itself already makes the importance clear. +which is unnecessary when the writing itself makes the importance clear. ### The "Almost" Hedge Models rarely commit to an unqualified statement. Instead of saying a pattern "always" or "never" does something, they write "almost always," "almost never," "almost certainly," "almost exclusively." The word "almost" shows up at high -density in model-generated analytical prose. It's a micro-hedge, less obvious -than the full hedge stack but just as diagnostic when it appears ten or fifteen -times in a single document. +density in model-generated analytical prose. It's a micro-hedge, diagnostic in +volume. ### "In an era of..." @@ -184,7 +176,7 @@ times in a single document. A model habit as an essay opener. The model uses it to stall while it figures out what the actual argument is. Human writers don't begin a piece by zooming -out to the civilizational scale before they've said anything specific. +out to the civilizational scale. --- @@ -196,7 +188,7 @@ out to the civilizational scale before they've said anything specific. Every argument followed by a concession, every criticism softened. A direct artifact of RLHF training, which penalizes strong stances. Models reflexively -both-sides everything even when a clear position would serve the reader better. +both-sides everything. ### The Throat-Clearing Opener @@ -204,8 +196,7 @@ both-sides everything even when a clear position would serve the reader better. > has never been more important." The first paragraph of most model-generated essays adds no information. Delete -it and the piece improves immediately. The actual argument starts in paragraph -two. +it and the piece improves. ### The False Conclusion @@ -241,7 +232,7 @@ vague than risk being wrong about anything. > "This can be a deeply challenging experience." "Your feelings are valid." Generic emotional language that could apply equally to a bad day at work or a -natural disaster. That interchangeability is what makes it identifiable. +natural disaster. --- @@ -251,21 +242,20 @@ natural disaster. That interchangeability is what makes it identifiable. If the first section of a model-generated essay runs about 150 words, every subsequent section will fall between 130 and 170. Human writing is much more -uneven, with 50 words in one section and 400 in the next. +uneven. ### The Five-Paragraph Prison Model essays follow a rigid introduction-body-conclusion arc even when nobody asked for one. The introduction previews the argument, the body presents 3 to 5 -points, and then the conclusion restates the thesis using slightly different -words. +points, and then the conclusion restates the thesis. ### Connector Addiction Look at the first word of each paragraph in model output. You'll find an unbroken chain of transition words: "However," "Furthermore," "Moreover," "Additionally," "That said," "To that end," "With that in mind," "Building on -this." Human prose moves between ideas without announcing every transition. +this." Human prose doesn't do this. ### Absence of Mess @@ -276,8 +266,7 @@ a thought genuinely unfinished, or keep a sentence the writer liked the sound of even though it doesn't quite work. Human writing does all of those things regularly. That total absence of rough -patches and false starts is one of the strongest signals that text was -machine-generated. +patches and false starts is one of the strongest signals. --- @@ -289,7 +278,6 @@ machine-generated. Zooming out to claim broader significance without substantiating it. The model has learned that essays are supposed to gesture at big ideas, so it gestures. -Nothing concrete is behind the gesture. ### "It's important to note that..." @@ -302,8 +290,7 @@ verbal tics before a qualification the model believes someone expects. Models rely on a small, predictable set of metaphors ("double-edged sword," "tip of the iceberg," "north star," "building blocks," "elephant in the room," "perfect storm," "game-changer") and reach for them with unusual regularity -across every topic. The pool is noticeably smaller than what human writers draw -from. +across every topic. --- @@ -314,10 +301,9 @@ Humans write "crucial." Humans ask rhetorical questions. What gives it away is how many of these show up at once. Model output will hit 10 to 20 of these patterns per page. Human writing might trigger 2 or 3, -distributed unevenly, mixed with idiosyncratic constructions no model would -produce. When every paragraph on the page reads like it came from the same -careful, balanced, slightly formal, structurally predictable process, it was -generated by one. +distributed unevenly. When every paragraph on the page reads like it came from +the same careful, balanced, slightly formal, structurally predictable process, +it was generated by one. ---