iiinova — where your digital self keeps evolving

a wiki editor @aionwikipedia.bsky.social · 01/10/2026

anecdotally, I've noticed a pattern in recent AI output where it pedantically, compulsively hyphenates compound adjectives that even the most nitpicky proofreaders leave alone (source: am one) turns out openai has actually suggested a prompt to suppress this github.com/openai/codex...

100

a wiki editor @aionwikipedia.bsky.social · 01/10/2026

(yes, I have seen the needle article; imo it's looking at the wrong things, I don't know why people get so hung up on punctuation when there is much higher-signal stuff)

000

a wiki editor @aionwikipedia.bsky.social · 30/09/2026

in case you're wondering what the latest indicator of AI-generated text is, it's this

A screenshot of Wikipedia search results on three unrelated topics, all of which contain a near identical sentence about how a claim should not be treated as fact without further sourcing

143

a wiki editor @aionwikipedia.bsky.social · 29/09/2026

pretty good piece on the ubiquity of AI-generated writing notably, every single person the author asked about their flagged-for-AI writing confirmed that they used it, which is consistent with my experience as well www.businessinsider.com/ai-detector-...

businessinsider.com

I used an AI detector for a month. What I found from my peers, dates, and heroes horrified me.

How much of our writing should we cede to AI? Does it even matter? I spent weeks running everything I read through an AI detector — here's what I saw.

120

Reposted by a wiki editor

Colin @colin-fraser.net · 28/09/2026

I already invented this it's called a fine

1393

a wiki editor @aionwikipedia.bsky.social · 27/09/2026

Wikipedia just added a feed to detect people copying and pasting from AI (not an "AI detector," it works off clipboard data) here's what it looks like after just 3 days en.wikipedia.org/wiki/Special...

en.wikipedia.org

Recent changes - Wikipedia

110

a wiki editor @aionwikipedia.bsky.social · 25/09/2026

person accidentally posts their AI conversation onto wikipedia, complete with a global green/amber/orange threat level map of which countries' wikipedias are in danger of blocking them cs.wikipedia.org/wiki/Jean_Br...

000

a wiki editor @aionwikipedia.bsky.social · 21/09/2026

why do people always hedge with "I used a tool"

Excerpt: In a follow-up statement about the composition of the email, Schnell wrote that he “developed the substance and wrote a draft” of the email, then “used an Al tool to assist with language and clarity, taking on several suggestions (including reworded sentences), before reading it through, making minor changes and sending it to you with my signature.” Highlighting the part about the AI tool

110

Reposted by a wiki editor

Colin @colin-fraser.net · 17/09/2026

In the post RLHF world this doesn’t amount to literally repeating its training data. But it’s completely sensible to think of it as ~repeating some hypothetical training data which is highly pleasing to the raters. And the raters love brass keys and protagonists named Mara.

1131

Reposted by a wiki editor

Colin @colin-fraser.net · 17/09/2026

When I say it’s a stochastic parrot, this is really what I’m talking about. I understand I am using the term slightly differently to how the original authors meant it. But what I am saying is: it generates text by bouncing around randomly in, often, a surprisingly small region of text space.

2123

Reposted by a wiki editor

Colin @colin-fraser.net · 17/09/2026

these four "books" were "published" **today**

4337

a wiki editor @aionwikipedia.bsky.social · 18/09/2026

Example of how many "AI tells" are obvious yet so hyperspecific most people wouldn't notice: There's one person on Wikipedia who spends their time fixing articles where people messed up the categories in one way or another. They've been doing this for over 12 years and have over 1.7 million edits.

110

Reposted by a wiki editor

Vauhini Vara @vauhinivara.bsky.social · 16/09/2026

I profiled the startup behind Pangram, the AI detector everyone’s talking about. It works really, really well. I wanted to understand how they did it. and also: why people still don’t trust them. (also, news: Penguin Random House is using it.) gift link! ✨

bloomberg.com

The World’s Most Reliable AI Detector Has a Human Problem

Armed with a near-perfect detection tool, Pangram is fighting to convince the public AI cheating can be caught.

68342

a wiki editor @aionwikipedia.bsky.social · 11/09/2026

visualization about the amount of (known) AI slop on Wikipedia versus the amount that people have been able to clean up dvrted.github.io/AINB-stats/

Chart of articles on the Wikipedia AI noticeboard over time from the past few days, September 7 to 11. The number of articles to cleanup has risen from about 21,000 to about 24,500. The amount that has been done has held steady.

121

a wiki editor @aionwikipedia.bsky.social · 10/09/2026

guy who develops AI psychosis except for ice "an ice age can cause an extinction event so we have to consider the possibility that ice is alive"

010

Reposted by a wiki editor

Colin @colin-fraser.net · 05/09/2026

kind of beside the main point but, what? What artificial intelligence analysis? By who? You, the reporter? Do you know how to conduct such an artificial intelligence analysis? I thought that reporters were more like supposed to report on what experts say, not just prompt Claude themselves or w/e

510717

a wiki editor @aionwikipedia.bsky.social · 28/08/2026

a fun fact about wikipedia is that the ai cleanup noticeboard (the page where people coordinate desloppifying articles) is now so large that it regularly crashes the browser if you try to load or edit it

000

a wiki editor @aionwikipedia.bsky.social · 27/08/2026

the thing about the "people are just saying 'clanker' because they want to say a slur" discourse is that we already know what people do when they want to say a slur without saying it, and it's stuff like "highly regarded" (or other algospeak) or references to pre-existing chud memes

100

Reposted by a wiki editor

John Herrman @jwherrman.bsky.social · 26/08/2026

maybe I'm being obtuse here, but I really get hung up on labor equivalence measures like this: it's a little bit horsepower, a little bit energy slave, a little bit man-hour, a little bit kilogirl (the computing output equivalent of 1000 women)

5384

a wiki editor @aionwikipedia.bsky.social · 22/08/2026

we're in a weird place with AI detection where the discourse has bifurcated into "this only applies to GPT-4" and "this only applies to Claude" the latter is an improvement since it's not 3 years out of date, but Claude has a really low market share (though probably overepresented by substackers)

110

a wiki editor @aionwikipedia.bsky.social · 19/08/2026

you don't actually have to run your own writing against AI detectors. you know what your writing process is. the time it takes painstakingly tweaking wording over and over and watching the little number move or not is time better spent doing things that actually improve your writing

101

Reposted by a wiki editor

Colin @colin-fraser.net · 19/08/2026

Here are some facts about "watermarking" LLM text. It's impossible to watermark text in a way that would perfectly discriminate between AI and non-AI text because that is just not a function of the text. A person or an LLM can conceivably produce any text. You can only make probabilistic guesses.

23510

a wiki editor @aionwikipedia.bsky.social · 13/08/2026

here's how modern LLMs do "it's not X it's Y" by the way (there are 22 instances of it here, one paragraph alone has 4)

a lot of claude slop not worth copying here except that in 3 lines there are two negative parallelism constructions with the phrase rather than

110

a wiki editor @aionwikipedia.bsky.social · 09/08/2026

credit where it’s due: regardless of your thoughts on the economist in general, this article about AI-generated writing is excellent www.economist.com/culture/2026...

economist.com

How to spot AI writing

Large language models like long words and em-dashes—or do they?

000

a wiki editor @aionwikipedia.bsky.social · 05/08/2026

more analysis on Grokipedia and its propensity to use science-coded-if-not-actually-science wording for everything (also on its Elon Muskness, of course)

000

a wiki editor @aionwikipedia.bsky.social · 02/08/2026

not quoting as to not blow up their spot, but the claim that people who are good at identifying AI writing are "doing critical text analysis for the first time in their adult lives" is deeply insulting, both on a personal level (copyeditor for 15+ years) and as a general assumption of bad faith

110

a wiki editor @aionwikipedia.bsky.social · 29/07/2026

"These detectors are notoriously, wildly inaccurate,” Alice Lemee, a ghostwriter and digital writing coach, said in a video she made about Substack’s new AI detection features." that's so weird, I can't believe you'd get flagged for AI

Public LinkedIn post by Alice Lemee: Welcome to Confessions of an AI Burnout 💌

When I turned 29 in September, everything was going great. I was on path to hitting a 💰 milestone in my freelancing business. I'd just moved into an apartment in NYC. I was doing the "adult" thing.

But by that January, my income tanked 60% and I felt lost as I'd ever been.

What happened?

It was (partly) because I'd bought into AI's siren call. It made my work "easier..."until it didn't.

I started to feel my brain–and writing skills–atrophy.

110

a wiki editor @aionwikipedia.bsky.social · 24/07/2026

some of the subtler elements of AI writing are most easily seen in copyedits: - "provide" turning into "aim to enhance student success by providing" (injecting promotional slop and turning it into a participial) - blandifying by turning "risky" to "harmful" (which also subtly changes the meaning)

Paragraph written without AI: School-based prevention programs provide school-wide education, skills training and supports intended to improve student success in school by teaching interpersonal and communication skills, increasing self-awareness, and reducing risk factors that lead to risky behaviors.

An AI rewrite of a paragraph that inserts slop like the phrase aim to enhance

110

a wiki editor @aionwikipedia.bsky.social · 23/07/2026

relatedly, this is an absolutely hilarious press release "While specific details about our history and unique offerings are limited in publicly available sources, our commitment to excellence drives us." www.workventure.com/company/mce-...

workventure.com

000

a wiki editor @aionwikipedia.bsky.social · 23/07/2026

new Randi Prize: $1000 to anyone who can produce an example of a paper they wrote in high school, prior to 2023, containing a sentence like "Comprehensive biographical details, assessments of impact, or documentation of major individual achievements remain limited in publicly available sources."

000

Reposted by a wiki editor

Morgan Clendaniel @mclendaniel.bsky.social · 23/07/2026

Seeing a simpler version of this in people's ability to detect AI writing: people who can clock ChatGPT tone are getting eaten alive by Claudeslop because they never talked to Claude and can't hone in on it the same way (and it's much more convincing and compelling than they think AI can be).

271

a wiki editor @aionwikipedia.bsky.social · 23/07/2026

I guess "catch a 5th grader using :contentReference[oaicite:1]{index=1}" was a bit hard to fit into the ad space

010

a wiki editor @aionwikipedia.bsky.social · 21/07/2026

commentariat: "AI writing just sounds like good writing! do you just hate good writing?" AI writing: "A 2025 City of Mississauga report about parking-related zoning amendments lists Westdale Mall among several properties proposed for a "Retail Centre" definition in the city's zoning by-law."

100

Reposted by a wiki editor

Wiki Education @wikieducation.bsky.social · 14/07/2026

Our latest data shows that the percentage of our participants' Wikipedia articles flagged as being written by AI has decreased since we began interventions in mid-2025. Explore how we handle generative AI usage, why verifiability remains a top concern, and what comes next: wp.me/p4LBez-XMK

wp.me

Generative AI and Wikipedia: The last six months

In January, Wiki Education published a widely read blog post about our experiences with Wikipedia editing and generative AI chatbots like ChatGPT, Claude, Gemini, and Copilot. It sparked lots of co…

044

a wiki editor @aionwikipedia.bsky.social · 14/07/2026

what people sound like when they say they don't want to write/code/whatever without ai youtu.be/gWGTehbT2LQ?...

youtu.be

I have to have my Tools (It's Always Sunny In Philadelphia)

YouTube video by bloodsploosh

000

a wiki editor @aionwikipedia.bsky.social · 11/07/2026

isn't it wonderful to live in a world where you can actually check stuff rather than making glib assumptions

110

Reposted by a wiki editor

John Herrman @jwherrman.bsky.social · 09/07/2026

The AI industry is prone to massive, frequent, and intensifying narrative mood swings. Part of this is typical of a boom. But I think it's also intrinsic to the tech itself nymag.com/intelligence...

There are plenty of reasons for this pattern. We're going through a massive investment cycle in a new technology, which understandably produces questions about bubbles, polarizes the economic discourse, and invites trader-brained, boom-or-bust interpretations of new developments. The economic stakes are now unavoidably high, and whatever happens is genuinely everyone's problem. Relatedly, while there's plenty of coverage of Al that tracks and contributes to these swings, the dominant mood of the industry - up to and including the communicative styles of AI CEOs
— is substantially determined in the tech hothouse of
X, where it's happening and nothing ever happens appear to be the only available positions, or at least the only ones that get engagement. At the same time, fears of job loss, talk of automation, the visibility of obvious externalities like slop content and cheating in school, and emerging worries about data-center construction have given tech's latest investment cycle an unusually powerful ethical and political dimension and turned the potential success or failure of the broadly defined object of AI into mainstream rooting interests.

For one, despite its obsession with forecasts and timelines, this is an industry without a good predictive theory of what its product will soon be good at. It wouldn't be unfair to say that the industry's actual approach to figuring out where things are going is to train a model and see, which gives AI researchers, and their companies, a genuinely unusual relationship to their products: They're building them, sure, but they experience their actual capabilities
- like a sudden aptitude for advanced math after a long stretch of profound innumeracy — as something similar to discoveries. This was clear in the way that Anthropic talked about the cybersecurity capabilities of Mythos as a strange, emergent phenomenon of training general-purpose models. (Contrast that with the development cycle in the nearby and similarly high-stakes semiconductor industry, for example, where swings in sentiment are comparatively mild or at least further apart.)

33811

a wiki editor @aionwikipedia.bsky.social · 08/07/2026

this can happen in even subtler ways. example: someone's "Grammatical fixes throughout the article and rephrasing the article to have a more neutral and clear tone" (later confirmed as AI) changed the text "[name] did not support Donald Trump" to "[name] withdrew her support for Donald Trump"

030

a wiki editor @aionwikipedia.bsky.social · 08/07/2026

interesting report but kind of lays bare how differently AI presents in different contexts, e.g., fictional manuscripts vs. nonfiction; a lot of the stuff that indicates AI usage here doesn't apply outside the fiction world (and vice versa)

211

a wiki editor @aionwikipedia.bsky.social · 02/07/2026

directionally correct, but could they have picked a different dude

my ai compass test results, pretty much at 1, negative 5, labeled The Skeptic, named after someone I wouldn't be surprised to learn searches their name, with the description: It's oversold, it's losing staggering amounts of money, and the harms are arriving faster than the value. You've done the math on the unit economics and it doesn't close. You are loud about this, prolific about this, and you suspect history will vindicate you.

120

a wiki editor @aionwikipedia.bsky.social · 21/06/2026

Harper's Bazaar has apparently joined Granta in pronouncing what appears to be an AI-generated short story the winner of its 2026 Short Story Competition:

Nabeel S. Qureshi on Substack: Another apparently AI-generated story wins a literary prize, this time judged by a panel including the novelist Ruth Ozeki.

Literary prizes need to start including Pangram checks in their process, or else change the rules to make AI writing ok. It’s very simple!

230

a wiki editor @aionwikipedia.bsky.social · 20/06/2026

putting AI summaries on articles in The New Yorker is just such a spectacular misunderstanding of who reads the publication

010

a wiki editor @aionwikipedia.bsky.social · 17/06/2026

read the study behind the "it's really easy to manipulate AI with tiny Wikipedia edits" and the wiki angle here is a little overstated; this particular attack sticks out a lot more in a wiki article than, say, a reddit thread asking for recommendations, and gets zapped fast arxiv.org/pdf/2605.24245

arxiv.org

100

a wiki editor @aionwikipedia.bsky.social · 12/06/2026

fun fact about Grokipedia: edits will get declined due (supposedly) to its language cutoff, which turns out to be a problem for a real-time encyclopedia

Grok output from Grokipedia, generated on April 4, 2026, declining an edit request to add information about the 2026 Iran War because it isn't 2026 yet

100