Notices by Kathy Reid (kathyreid@aus.social)

Thought for the day:

We used to measure data centres in how many rack units of capacity they had - e.g. this data centre has 5000 rack unit equivalent capacity. This is a measure of physical capacity.

Then we measure data centres in TFLOPS - how many trillions of floating point operations they could perform. This is a measure of computational capacity.

Now we measure data centres in Gigawatts - how much power they consume. This is a measure of power consumption, not physical space or output.

What *should* we measure data centres in? Gigalitres of water consumed? Households disrupted by their construction?

#data #dataCentres #HPC

When the USA descended into civil war, I didn't think that the trigger would be .. access to vaccines.

My Mum has a new #corgi puppy called Holly. She is 7 mths old, adorable, playful and an absolute joy.

I have taken to referring to her as Her Hollyness. This is both blasphemous and accurate.

#corgisofmastodon #corgi #HerHollyness

It started slowly.

Seattle-based companies made the move to Vancouver, with generous re-location packages.

The Valley, bit by bit, moved northward, first capitalising on the cheaper real estate in Portland and Seattle, then joining the ex-SFats in Canada.

---

SoCal had crumbled a decade prior - the nexus of water-guzzling, power-sucking AIs that mimicked Hollywood actors only worked as long as there was water and power - and Hollywood actors to mimic.

The LA fires of '32 had destroyed so much infrastructure the city was broke - and dry.
And once they'd been cloned and recast as AI avatars of themselves, actors retired - on albeit reduced residuals. No-one flew to LAX with a dream and a cardigan. Those left were either ICE agents or Undocumented they hunted.

---

The Big Beautiful Bill had exactly the intended effect - an unbridgeable chasm between rich and poor. You either had capital or you died - usually from something preventable like gum disease or pneumonia.

Those with capital moved north.

Those who only had labour - human, devalued yet expensive, connective yet computed labour - stayed. Inflation grew. The $USD sank.

The future, as the Great Dismal wrote, was already here - it just wasn't evenly distributed.

Five eyes became three when Australia and Aotearoa ripped up ANZUS in '30. Canada followed and Starmer, enjoying an unprecedented fourth term in a time of turmoil, made it one in '31.

---

NoCal took a little longer - the vulture capitalists hanging on to the carrion of would-be unicorns, hoping to get a RoI.

After adding AI to everything, they had to enshittify to get a return. Inferencing a 5Q parameter model doesn't scale, even with a Redis cache the size of Sonoma county.

AI had, paradoxically, re-ignited the Open Source movement, who, determinists at heart, rejected the stochastic musings of the Bullshit Machines.

A team from Eleuther paired up with Wikipedia and dropped a reciprocal-licensed distilled RAG model that inferenced quicker than DeepSeek. They might have had help from CC Signals, who reached out to people interested in opening their information - but not for profit.

Then they used BitTorrent to do results caching. Inference cost dropped to marginal. A couple of big names saw, ahem, a slight decrease in market cap.

---

Anyway, enough of the history lesson!

We're delighted to welcome you to Victoria Island, the new headquarters of FAANG, Inc.

We've pivoted a little in the last few years, and now most of our work involves geo-engineering so we can stay under 3 degrees.

We like it here, and we hope you do, too.

#TootFic #AlternateHistory #MicroFic #MicroFiction

Are you a top US-based scientist looking to relocate to #Australia?

The Australian Academy of #Science is now offering generous relocation packages and guaranteed #research funding.

https://www.science.org.au/news-and-events/news-and-media-releases/are-you-a-top-us-based-scientist-wanting-to-relocate-to-australia

#FediScience

:boost_requested:

@inthehands

Is it metathesis?

https://en.wikipedia.org/wiki/Metathesis_(linguistics)

Hey Spotify, as someone who lost my ovaries to cancer at aged 33, I would rather not have egg freezing ads, plskthx

"I'm sorry, Professor, but unfortunately you haven't met your monthly KPIs for three months, and the rejection from NeurIPS was the deciding factor. Our Corporate UNattaching Teams will be in touch soon with next steps."

She sighed, more resignation than shame.

---

It started before the decimation of science in '25. Academic publishing had corporatised long before that; journals were a cash cow built on free labour - from writers and editors.

When the Institutes lost funding in '25, universities were forced to rely on their endowments. The Ivy Leagues survived a few years - but by the early 30s, when Trump was in his third term, they too were broke.

And the Journal-AI Conglomerates stepped in.

Wiley-Taylor-Francis-Anthropic, Elsevier-Springer-Gemini, SAGE-OpenAI.

The journals had each merged with key generative AI companies in the late 20s. The prevalence of AI slop and the Token Crisis meant that there no more human-created tokens to feed the LLMs ... except those in paywalled journals.

The AI companies got their tokens, and the journals got their token lucre. It was a match made in hell.

This was all by design, of course.

---

"So, tell me how your new h-index v2 works, and how my h-index v2 means that I don't meet my KPIs?"

"Ah, Professor, we've changed our ranking algorithm, and it no longer weights the h-index v2 as heavily. It's much more focused on how many words are written and published."

"So you can harvest them for tokens?"

Her interlocuter cleared his throat. She wasn't sure it was a "he" specifically - the voice AIs were now indistinguishable from humans, but she suspected he was a fallible meat sack.

---

Each of the Journal-AI conglomerates started to buy up universities - like the health insurers had done in the 2010s and 2020s - forcing up the price of degrees and adding barriers to academic publishing.

They'd brought pressure to bear on the US government - it had been easier than they thought - to ban Open Access - casting it as a tariff on US-grown research.

Then they simply enshittified.

Academics were given publishing KPIs - essentially told to create research as a front for generating human tokens, to be fed into ever-larger LLMs.

Great research was no longer the point - tokens were.

---

1/2

#Tootfic #Microfiction #HigherEd

If you watch one thing today, watch this impassioned, brilliant, incisive talk from Carole Cadwalladr for TED Conferences.

"It's a coup."
"A coming autocracy."
"A type of power the world has never seen before."
"The broligarchy doesn't want you to have a private life."
"We are already living inside the architecture of totalitarianism."

This. This is what a digital coup looks like.

Thank you Carole for your bravery. For leading the resistance.

https://www.youtube.com/watch?v=TZOoT8AbkNE

My brain works in very funny ways. I was listening to a series that had some Korean dialogue, and then I heard Mmmbop - the track from the 90s - on the radio - and now it's playing in my brain but the lyrics are "gimbapp mmm gimbap" and now I want sushi.

Excellent piece from @zephoria from Tech Policy Press arguing for "interventionism" - a form of pragmatism that seeks to recognise the uncertainty in socio-technical systems while ascribing importance to legal, policy and social mechanisms that counter-balance techno-solutionist narratives.

#STS

https://www.techpolicy.press/we-need-an-interventionist-mindset/

I didn't expect "using a Torrenting client to seed US government scientific datasets because the policy settings are erasing years of research" on my 2025 bingo card but here we are.

You wouldn't download America's scientific research output would you?

#SafeGuardingResearch

Lovely piece from #UVA Data Science where @histoftech talks about early computer dating apps.

Super interesting to learn about their cultural and social place at a time when match-making - whether by humans or algorithms - was stigmatised.

https://datascience.virginia.edu/news/ibm-hinge-professor-mar-hicks-history-computer-dating

If a company employs you because of your knowledge of byte pair encoding, does that make you a token hire?

#ML

IDEA: I have an idea for an art installation which is collections of #OpenSource conference badges as political artefacts, showing transitions over time between, e.g. BarCamp DIY-style name tags with Sharpie, to more corporatised conferences, e.g. OSCON, to grassroots conferences, e.g. #FOSDEM, pulling at threads such as the self-decoration of badges, normalisation of pronouns on badges, sponsor relationship to badges and conf design.

What would those narratives look like?

Day 2 of #EverythingOpen with @itgrrl talking Everything Open All At Once and relating everything open to the cultural phenomenon of EEAAO - and "what ifs".

I have to admit I'm pretty annoyed at getting asked to donate to @wikipedia when I contribute content and edits, and when companies like OpenAI, Anthropic, Microsoft, Google et al scrape it to build their #LLMs then sell that back to us.

The Mozilla #CommonVoice #dataset v20 was released yesterday - the largest open #speech dataset in the world. My #dataviz, linked below, shows a continuation of patterns seen for some years now:

➡️ There's more data collected for #Catalan (ca) than for #English (en) - testament to the independence and language reclamation efforts in Catalunya. Language and cultural transmission are deeply intertwined.

➡️ Some of the newer #languages to Common Voice, like #Ligurian / #Genoese (lij) have contributions from mostly older speakers, which is unusual in comparison to the rest of the dataset. This may reflect the population that currently speak those languages - as many regional languages in Italy are in rapid decline.

➡️ Some languages such as Eastern Mari / Meadow Mari (mhr) - a #Uralic language spoken in the Mari-El Republic within Russia - have samples from predominantly female-identifying speakers, again contrasting to the rest of the dataset. Other languages here include #Cantonese (yue), #Georgian (ka), and #Kalenjin (kln).

➡️ A key part in the preparation of the Common Voice dataset is the validation of utterances to assure they match their written transcription - which requires at least two validations by separate speakers. Some newer languages to Common Voice, such as Erzya (myv) and Moksha (mdf), both Uralic languages, have nearly 100% validation.

What are your interpretations of the dataset?

https://observablehq.com/@kathyreid/mozilla-common-voice-v20-dataset-metadata-coverage

Most of the non-paywalled web became unreadable about 2021.

Stuffed with bevel to bevel ads that even a #PiHole or #uBlock struggled to swat away, reading long-form content became a chore.

Just like television, where the volume on ads mysteriously doubled, web ads became shoutier, and animated, popping up over the real content payload, diverting reader attention, increasing cognitive load and raising blood pressure.

"Subscribe for an ad-free experience!", the sites beckoned. "We're really Nice Guys when you get to know us".

Except the subscriptions were still full of ads, just less shouty, and more targeted to the plus-addressed email you used to subscribe.

This had an unusual and unexpected effect.

It trained people to focus their attention on long-form content. Moulded their elastic, dopamine-addled brains to seek out the nuggets of pure and informational expression among the wasteland of attention-sucking weeds. Sharpening their retention, developing the stamina to persist while resisting distraction.

They were, unintentionally, growing grad student brains.

By 2026, doomscrollers had discovered arXiv, probably when WIRED stuffed one more ad in a profile of Martha Wells for the latest Murderbot and an Aussie sci-fi fan rage-tooted on the Fedi.

Suddenly, the Discourse shifted.

Why struggle through a 1000 word op ed with thirty seven ads for - I forget what they were because my brain ignored them - when you could read a 9k word review on the current state of #LLM evaluations, unencumbered by eyeball-seeking cruft?

Doomscrollers started on arXiv then instead of getting frustrated, moved to OpenReview, processing their thoughts on what they'd read not in the form of likes or emoji but as peer reviews.

The ad industry gasped one final choking breath, after having been denied access to arXiv or OpenReview.

#Tootfic #Microfic

After decades of being "nice" and a "people pleaser" it's been really interesting to see how people respond to me setting much tougher personal boundaries.

The level of entitlement to my time, energy and attention that some folks have is unbelievable. It's definitely been a learning journey.