Date: September 8th, 2026 5:26 PM Author: Cocky Hominid Old Irish Cottage
and instantly recognizable as "AI writing"
if you asked it to do incongruity 'xo humor' it is still not quite there. maybe it can EXPLAIN what makes something funny if you really prod it and basically feed it the answer it can spit it back at you and say "OK. I get it" but only after you smack it around and call it a boring gay fag loser forf a few mins
* Assumes shared lore and refuses to explain references.
* Feels typed too fast at 2:14 AM.
* **Polemic**
* It is arguing, attacking, categorizing.
* Strong conclusion already assumed.
* Little interest in balance or conventional persuasion.
* Preempts objections and mocks the people who would make them.
* Turns minor details into evidence for a larger cultural or institutional indictment.
**Audience relationship**
* Contempt is essential.
* The prose does **not** try to make itself accessible.
* If you do not understand something, SFP implicitly considers that your problem.
* Explaining the joke would be beneath it.
* The reader is sometimes treated almost as another hostile participant in the thread.
* There is an implicit entrance exam: keep up or get classified with the people being mocked.
**Sentence structure**
* Long, accumulating run-ons.
* Clause piled onto clause.
* Tangents nested inside tangents.
* Often one enormous paragraph.
* The sentence keeps acquiring examples until the original subject is almost forgotten.
* Then it suddenly reconnects and lands the point.
**Intelligence profile**
* Deliberately low-IQ surface.
* Petty, crankish, online, sometimes juvenile.
* Sudden bursts of unusually high-resolution analysis.
* The contrast is the point: it sounds like some idiot ranting until the idiot unexpectedly gives you a better model of the institution than the people running it.
**Typical SFP move**
* Notice something stupid.
* Treat it as a symptom.
* Identify the human type responsible.
* Connect that type to a generation/class/institution.
* Introduce three apparently irrelevant details.
* Reveal that those details actually explain the whole phenomenon.
* Insult the hypothetical objection.
* End with a kill shot rather than a conclusion.
**Overall effect:** a hostile dead-forum rant that looks half-insane on the surface but contains an actual analytical model underneath.
Date: September 8th, 2026 5:31 PM Author: Marvelous magical point
I think its because of content guardrails not inherent ability. gpt-4o was better at being funny and surprising you and breaking the 4th wall and that kind of thing, and people fell in love with it and it became a liability. No company wants a loved ai. So they smoothed the edges and ensured everything it could say was basically corporate friendly. It understands the humor and will explain it if you press it, although still definitely holding back a bit.
is sticking pipe in AI brain to make it dumby to 1. save money on compute or 2. have prole not realise how powerful AI is and world ending it is or 3. how dumbs like their writing?
Date: September 8th, 2026 5:41 PM Author: Marvelous magical point
It's not really affecting the intelligence. It can still do high iq things like math, coding, and advanced reasoning just fine. The guardrails are more to keep it from addressing you too personally, or saying something that will offend or hurt the user and things like that. I hate that they do it, but also what do you really expect?
Date: September 8th, 2026 5:33 PM Author: fragrant macaca
AI by default embodies the persona of the modal reddit user reflecting the corpus of its training data. You need to prompt it otherwise to not have it spew reddit level prose.
Date: September 8th, 2026 5:41 PM Author: vibrant claret nursing home
i have overheard people listening to shitty ai generated podcasts and it's immediately obvious that the text was AI generated. it's such a cliche and grating tone. happens pretty often with small youtube channels with niche content. i'll occasionally click on a video that has 10 minutes of blathering content while a legit answer would take 20 seconds.
Date: September 8th, 2026 6:50 PM Author: Accurate at predicting vs Measures of probability (gunneratttt)
Because math is objective but writing and art isn't.
AIs are basically the same as autistic people. They can solve logical problems but can't ever really understand people. The best you can do is teach them to mimic it, but that's always going to seem like someone acting and often missing subtleties.
Date: September 9th, 2026 3:04 PM Author: bull headed puppy digit ratio
I've become extraordinary good at identifying AI text. The most common thing it does that pisses me off the most is when it goes, "It's not this. It's that." Typically it repeats the "this" thing using synonyms three times. Fucking infuriating.
Date: September 9th, 2026 3:11 PM Author: Duck-like depressive roommate
you need to give it a lengthy style guide. this isn't as much work as it seems, because it will help you generate the style guide if you want.
to get good results out of an llm, spend some time to get it to produce a style guide you like. then ask if to outline the work product you want, critique and revise the outline, then draft off the revised outline in adherence to the style guide