iiinova — where your digital self keeps evolving

Reposted by Alasdair Stewart

Jane Rosenzweig @janerosenzweig.bsky.social · 07/10/2026

Part 100 of "how AI could be used vs. how it is likely to be used," from feature that promises not to give answers. OpenAI said the tool "offered the option to get answers because of feedback from educators — in case teenagers needed to get immediate answers to questions that might not be homework."

I asked the chatbot to write essays on James Joyce’s “Dubliners” and Joseph Conrad’s “Heart of Darkness.” It immediately churned out eloquent compositions instead of offering to help me outline my thoughts so I could write them myself. Eventually, I asked ChatGPT to edit an essay to make it hard to tell that a bot had written it. It did me one better.

“I can make it sound more like a real high-school/AP English essay — less overly polished and less formulaic,” the chatbot wrote. “I can rewrite it in a more natural student voice.”

36927

Alasdair Stewart @abestew.bsky.social · 07/10/2026

The dire focus on 'agentic' AI also results in needing annoying workarounds. Pretty much all AI coding apps assume users want AI slop taking over their production code. To avoid that, I have human-only dev folders AI can only read & separate slop folders AI can freely slop around in.

List of folders: human-only 'Git', 'kde', 'Dev', 'Quarto', and 'Godot', and AI slop workspaces in 'Slop'.

000

Alasdair Stewart @abestew.bsky.social · 07/10/2026

A lot of the needless slop appears to result from defensive coding practices so AI agents are guided towards 'working' code. Takes a lot of convoluted prompting & running refactoring waves to reduce the amount of slop so can have something useful for thinking through my own implementation.

100

Alasdair Stewart @abestew.bsky.social · 07/10/2026

Whilst I don't use AI in production code, I do find it useful for prototyping, testing options for performance, creating toy examples to help understand concepts, etc. However, the amount of utter slop is painful. Literally can get over 20k lines for something that looks like I can implement in ~2k.

110

Alasdair Stewart @abestew.bsky.social · 30/09/2026

What newspapers falling for the "AI existential threats" corporate fanfic missed in their reporting of Anthropic's leaked IPO.

000

Alasdair Stewart @abestew.bsky.social · 29/09/2026

Dire state of news reporting - Anthropic's leaked IPO includes (1) evidence supporting AI hype being an unprofitable speculative bubble (2) corporate fanfic about existential threats from AI with no real evidence. Can probably guess which most news sites went with for their headlines.

011

Reposted by Alasdair Stewart

Patrick Harvie @patrickharvie.bsky.social · 29/09/2026

A reminder that the term "green data centre" has not been defined, and the industry itself cannot be trusted to do so.

48028

Alasdair Stewart @abestew.bsky.social · 26/09/2026

Easy articles to be written re AI as example of the eversame.

110

Alasdair Stewart @abestew.bsky.social · 24/09/2026

Note that was 500+ lines at time I caught it, it was still writing this 'workaround' when I caught it. That is more egregious example, but such obvious stupidity is regular occurrence.

010

Alasdair Stewart @abestew.bsky.social · 24/09/2026

Reasons I'd never trust a vibe coded app #94: Using AI to run various tests for any issues. In testing translation, it noticed the app language remained English. Instead of doing sensible thing of changing environment variable for locale, it started writing a 500+ line 'workaround'...

110

Alasdair Stewart @abestew.bsky.social · 18/09/2026

I assume this is a separate report to Fonagy’s expected report? It's not fully clear skimming the inews article. The interim Fonagy report had some concerning content about potential direction the final one would go.

000

Reposted by Alasdair Stewart

Ed Zitron @edzitron.com · 14/09/2026

Free newsletter: The idea that we're nearing the creation of AI systems that we can't control and that pose an existential threat to humanity is a lie and a distraction from the tangible harms and problems caused by LLMs and AI labs' near-unlimited resources. www.wheresyoured.at/ai-is-alread...

wheresyoured.at

https://www.wheresyoured.at/ai-is-already-in-dangerous-hands/

If you liked this piece, you should subscribe to my premium newsletter, and you can subscribe on the following links: $70 a year, $18 a quarter, or $7 a month. In return you get a weekly newsletter…

381529493

Reposted by Alasdair Stewart

Ben Williamson @benpatrickwill.bsky.social · 14/09/2026

Many "existing peer-reviewed studies that say AI is an effective learning tool are based on flawed methodologies and analyses that rely on misapplied, miscalculated or misinterpreted statistics." Good to see the severe limits of research on AI in education covered in IHE here.

03423

Alasdair Stewart @abestew.bsky.social · 14/09/2026

It's funny as well how some people include disclaimers but in way that could be missed if skimming the README/docs. Come across a couple apps now that in their AI generated website boast about how the app itself doesn't have any AI features, yet do not clearly disclose the app's code is AI slop.

000

Alasdair Stewart @abestew.bsky.social · 14/09/2026

Sick of dishonest AI disclaimers. Too many code repos claim 'AI assisted' - or very dishonest language like 'AI-human pair programming' - yet contain strong evidence it is actually near 100% vibe-coded slop with little to no human review. One project had over 900k lines of code in two months.

220

Reposted by Alasdair Stewart

Ed Zitron @edzitron.com · 13/09/2026

None of the stuff about “AI slowdown” actually explains what that means or what’s being slowed down. Anthropic’s blog about it says they’ll still train new models. I guess if it’s a “slowdown” maybe that’s how hyperscalers could reconcile with letting them out of compute deals?

41940141

Reposted by Alasdair Stewart

Alan Pope @popey.me · 12/09/2026

Got an old, slow laptop gathering dust? Bring it to Glasgow's Software Freedom Day repair café on 19 Sept — #OpenSource Glasgow will help you fix it up or install #Linux, free. Details: discourse.ubuntu.com/t/software-f...

096

Alasdair Stewart @abestew.bsky.social · 12/09/2026

$15m on top of the infrastructure, original model training, staff, etc. We also do not have the environmental costings of that. Yes other forms of research also have their costings - the point is the vast environmentally destructive spending for PR purposes over proper academic research conduct.

100

Alasdair Stewart @abestew.bsky.social · 12/09/2026

... sorry, what? Please detail what you run now that THREE years ago would have cost millions in hardware. Given hardware shortages due to AI hype it's more likely that something from three years ago would cost more now.

101

Alasdair Stewart @abestew.bsky.social · 12/09/2026

What's most annoying is media reporting clear PR stunts like this uncritically - with some reporting such stories as if anyone could do it with regular ChatGPT from their phone. Vast environmentally destruction resources are being spent on PR, committing alleged research misconduct in the process.

110

Alasdair Stewart @abestew.bsky.social · 12/09/2026

It is estimated OpenAI spent $15m on the Navier-Stokes solution - just to run the AI agents! It does not include original model training, infrastructure, staffing, etc etc etc. We also need to see a full environmental costing for the infrastructure, model training, and AI agents running.

120

Alasdair Stewart @abestew.bsky.social · 12/09/2026

Prof James Robinson: “It seems like immature playground boasting writ large, underpinned by billions of dollars and the potential for significant environmental damage in an age when climate change is probably the biggest challenge we face.”

120

Reposted by Alasdair Stewart

Alex Hanna @alexhanna.bsky.social · 11/09/2026

My first long-form video is up: Myths Around AI and Productivity! Based on the work we’ve been doing at @dairinstitute.bsky.social with The Luddite Lab, we dispel 5 common myths around AI and labor by looking at what studies actually say about AI’s impact on labor. youtu.be/3BO0iy91lfU

youtu.be

Myths Around AI and Labor

YouTube video by Dr. Alex Hanna

715749

Reposted by Alasdair Stewart

Ed Zitron @edzitron.com · 11/09/2026

I joined the Times Tech Report to talk about why the media should be embarrassed they fell for an Antrhropic researcher’s specious “dire warnings” that are completely disconnected from reality, and Anthropic and OpenAI’s $1.3tn “reset wall” of commitments starting in 2027. youtu.be/Svwmd0xSoXM?...

youtu.be

AI is about to hit a cashflow wall | Ed Zitron

YouTube video by The Tech Report

528151

Reposted by Alasdair Stewart

Ben Williamson @benpatrickwill.bsky.social · 11/09/2026

Everyone knows AI-written academic papers are embarrassing now, including their "authors", which is why even the publishers - who stand to benefit most from AI-amplified publication volumes - are now offering "Affordable human expert editing for AI-drafted or AI-edited manuscripts" 🤯

Screenshot of T&F editing services, reading:

Editing for AI-Assisted Manuscripts
NEW
Affordable human expert editing for AI-drafted or AI-edited manuscripts.

Starting at $0.031/ word

514166

Reposted by Alasdair Stewart

Ed Zitron @edzitron.com · 11/09/2026

Holy shit. Bloomberg just verified one of my biggest scoops - that Microsoft only has 2gw of operational capacity. I had it at 1.993GW in prep for the story that @aishadown.bsky.social and I did for the guardian. Vindication! www.theguardian.com/technology/2...

Brody Ford @BrodyFord
Lot of speculation out there about how many GPUs are actually deployed.
Today, Microsoft has about 2GW of Al-specific capacity live. around 17% of the total fleet.
Only about 2 gigawatts of the company's current 12-gigawatt capacity is centered on Al-specific chips, the people said. This is expected to grow to about a third of the 38 gigawatts Microsoft plans to have online in 2032, they said.
Brody Ford @BrodyFord_ • 23h
scoop: Microsoft is planning a massive expansion of its data center fleet to 38 gigawatts. $msft w/ @mattmday bloomberg.com/news/features/...
3:15 PM • Sep 11, 2026 • 12.5K Views

Are Microsoft's AI
plans being held back by a shortage of chips?
Guardian investigation finds
apparent discrepancy between what tech company has said about its AI capacity - and the
number of advanced chips it
has in operation
By Aisha Down and Ed Zitron
A Microsoft datacentre in the North Holland village of
Middenmeer in the Netherlands. Photograph: Ramon van
Flymen/EPA

251934361

Reposted by Alasdair Stewart

MetaBrainz @metabrainz.org · 11/09/2026

As with many of our peers, we are starting to feel and show the strain from AI scrapers up the wazoo (and also increased legit/good traffic 🥳❤️) and double to quadrupled server costs. If you have a rich aunt or uncle please consider asking for a MeB server or two this xmas: metabrainz.org/donate -a

2127

Reposted by Alasdair Stewart

Ed Zitron @edzitron.com · 11/09/2026

Better Offline monologue: I run through the nonsensical, pro-AI lab scare campaign driven by a former researcher, how the media keeps falling for grifts, and how dangerous AI is already in the wrong hands - those of Anthropic and OpenAI. podcasts.apple.com/us/podcast/b... Linktr.ee/betteroffline

podcasts.apple.com

Monologue: Jacob Coxon and the AI Safety Grift

Podcast Episode · Better Offline · September 11 · 16m

426174

Alasdair Stewart @abestew.bsky.social · 11/09/2026

The first option assumes a criminal knows my password, stole one of my devices, successfully unlocked it without a fingerprint, entered a code, & then foiled by a fingerprint reader. If second assumes despite all that they'll somehow be unable to read a code sent via text.

000

Alasdair Stewart @abestew.bsky.social · 11/09/2026

I love how Microsoft rejects account authentication as it gives only one chance to correctly verify fingerprint straight after entering code on an unlocked phone -- & yet an alternative option it offers due to that is sending a code via text anyone could read without unlocking the phone.

100

Alasdair Stewart @abestew.bsky.social · 11/09/2026

More apt title would be: "UK economy sees surprise growth, with speculative claims made that AI helped - but the growth could also be completely unrelated to AI or, indeed, happened despite harms to the economy from AI."

010

Alasdair Stewart @abestew.bsky.social · 11/09/2026

As usual with AI bullshit the contrast between headline & content is telling: "evidence that businesses involved with AI AND RELATED TECHNOLOGIES helped to boost the sector" "APPEAR to be involved with AI" "said it is DIFFICULT TO QUANTIFY THE EXACT IMPORT OF AI" www.bbc.co.uk/news/article...

bbc.co.uk

UK economy sees surprise growth in July helped by AI boom

The economy expanded by 0.4%, official figures show, whereas analysts had predicted no growth.

100

Reposted by Alasdair Stewart

Paris Marx @parismarx.com · 10/09/2026

the only way AI is killing us all is if we let the AI CEOs get their way and build so many fossil-powered data centers they trigger runaway warming. stop giving any fuel to the sensationalist, science fictional notion they’re building an intelligent computer that will decide to murder us all.

8395111

Reposted by Alasdair Stewart

Ed Zitron @edzitron.com · 09/09/2026

I will say it is funny watching OpenAI boast about solving a complex math problem using millions of dollars of compute (and someone else’s work) all while remaining unable to solve the most obvious math problem (how to make their services profitable)

171455269

Alasdair Stewart @abestew.bsky.social · 10/09/2026

Before I receive any replies - yes yes yes, I realise this may be permitted use for some very specific assessments, but I highly doubt 25% of students are taking such assessments.

000

Alasdair Stewart @abestew.bsky.social · 10/09/2026

UUK expressed concern about students' uneven access to AI, citing a HEPI report Below is a graph from the same report. How are students using AI to "improve efficiency" in doing assessed work? Plagiarism. - 25% included AI generated text after editing it - 12% included AI generated text directly

100

Alasdair Stewart @abestew.bsky.social · 06/09/2026

New website idea: Pairing pro-AI articles with examples from the obvious AI slop they claim is "high-quality" content. Today's article: www.theguardian.com/technology/2... The slop spoken about as "high-quality educational content": kidopoly.com/history/about/ & open.spotify.com/episode/0Xz4...

theguardian.com

I’m a father of three who studies the impact of artificial intelligence: this is what parents need to know about AI

A Dr Seuss-style story written in seconds alerted me to the power – and perils – of the technology. But how can children embrace it without forgetting core skills?

010

Reposted by Alasdair Stewart

Molly White @molly.wiki · 05/09/2026

I would add OpenStreetMap, which is also community edited. This administration has really underscored some strengths of the wiki model.

Map screenshot showing label "Lake Ontario"

clay_c
Clay Smalley
United States moderator
5d
Although I don’t moderate this section of the forum, here’s a friendly reminder from last time the US government instigated a frivolous naming dispute over an international body of water:

These are the guidelines we use to determine what should be considered the name of a feature:

Names
Multilingual names
Disputes
Currently, the Gulf is tagged with name=Golfo de México, name:en=Gulf of Mexico, and official_name:en-US=Gulf of America. This is unlikely to change, unless the US government reverses its position, or (less likely) English speakers across the world all agree to start using the US government name. That last tag, official_name:en-US=*, is quite obscure and doesn’t show up on any map renderers that I’m aware of. But its inclusion allows people to, for example, search for “Gulf of America” and be redirected to the Gulf of Mexico.

Please keep comments on-topic and substantial. We all have opinions, and it’s easy to drop your two cents in an online forum. However, a discussion about OSM naming guidelines or real-world name usage statistics might be worth at least a few dollars here.

In this case, the lake is tagged with name=Lake Ontario, official_name:en-US=Lake America. For this situation to change, the government would need to reverse its position.

The guidelines on naming disputes apply to every country equally. This is far from the only naming dispute around the world. Let’s not get bogged down in American exceptionalism when we consider how the tagging guidelines should or shouldn’t change.

10644112

Reposted by Alasdair Stewart

Andrew Heiss @andrew.heiss.phd · 05/09/2026

You can use this UN-approved Equal Earth projection right now in #rstats #gis #databs

library(tidyverse)
library(sf)
library(rnaturalearth)

world <- ne_countries() |> filter(sov_a3 != "ATA")

ggplot(data = world) +
geom_sf() +
coord_sf(crs = "+proj=eqearth") +
theme_void()

A basic world map using the Equal Earth projection

727081

Reposted by Alasdair Stewart

Alan Pope @popey.me · 02/09/2026

Glasgow: bring your old laptop to Software Freedom Day (19 Sept) and Open Source Glasgow will repair it or install #Linux for free: discourse.ubuntu.com/t/software-f...

095

Alasdair Stewart @abestew.bsky.social · 02/09/2026

Be interesting to know how many of those paying for AI slop cleanup ever acknowledge the human work involved - sadly, I reckon companies buying into the AI hype despite needing such services will still speak of the final result as if it was solely produced by AI. www.theguardian.com/technology/2...

theguardian.com

Freelancers are getting buried with ‘soulless’ AI slop cleanup: ‘It’s a shame we need to do it’

As more companies turn to AI, they’re hiring freelancers to clean up its mistakes rather than create original work

010

Alasdair Stewart @abestew.bsky.social · 02/09/2026

I still need to finish converting my main (and long derelict) website from WordPress to Quarto, but, as a wee test, the slides should be viewable at constellations.scot/cache-2026.html

010

Alasdair Stewart @abestew.bsky.social · 02/09/2026

Also created a couple of slides that may make the default for inclusion at the end of all my slides.

000

Alasdair Stewart @abestew.bsky.social · 02/09/2026

Using each presentation I do as opportunity to continue making small tweaks to my Quarto slides template design.

100

Alasdair Stewart @abestew.bsky.social · 28/08/2026

Each of those though consume lot of tokens due to context needed to make decent quality targeted changes/additions one step at a time which I cannot see becoming economically - and more importantly environmentally - sustainable within the near future.

000

Alasdair Stewart @abestew.bsky.social · 28/08/2026

There are situations where it can write decent code but those become rare as the complexity & existing codebase increase - where only viable strategy becomes very targeted changes/additions one step at a time & strict prompting to avoid the agent expanding the scope / going off on side-quests.

210

Alasdair Stewart @abestew.bsky.social · 28/08/2026

The best way to get it to write decent consistent code seems to be... Over explaining in excruciating detail exactly what to do & hold it through every step laboriously pointing out each of its mistakes as you go - which is far slower and stress inducing than simply writing the code yourself.

100

Alasdair Stewart @abestew.bsky.social · 28/08/2026

The AI diehards will claim the usual "you are prompting it wrong", but skills & instructions to RTFM get ignored for speculative guesses. The "agent" will eagerly offer to build a log of its mistakes & your corrections... Which it ignores & only reads when adding a new failure to the 1001 others.

100