iiinova — where your digital self keeps evolving

Zach Weinersmith @zachweinersmith.bsky.social · 16h

Personally, I wouldn't knowingly use AI-generated music even for the low stakes uses I have for background music (TTRPGs). I certainly wouldn't pay for it.

000

Reposted by Aaron Helton

One of my main mental buoys right now as ai eats the world is Bea Wolf 2. I don't think I can write any better than that, and it was hard, and every time I see a book get withdrawn (for secretly being ai) after winning many awards I think of the ballad of John Henry racing the steam drill*.

1420916

..recoup your losses with a few follow-up prompts. Most likely, though, you'll spend more time than necessary chasing a receding goal as the AI gets hung up on statistical minutia, invents new subtasks, and churns until you walk away or run out of tokens, all while it tells you how smart you are.

020

In gambling with money, you keep going when you're down in hopes of winning back what you lost. But the iron rule of that kind of gambling is that the house always wins. So you might get lucky, but on average, people don't. Likewise, you may craft a perfect one-shot prompt, or slightly more likely..

100

Also my experience more than half the time, which does little to dissuade me from my assertion that this is part and parcel of the gambling economics that seem to have captured commerce at the moment. Where each spin of the wheel offers the chance of correcting the error the first spin introduced.

Aaron Helton @aaronhelton.com · 17h

Yeah definitely looks to me like "throw it at everything and see what sticks well enough to call it a win", but even then, what counts as a "win" is qualified and might not even be durable.

020

I used to work in HR systems, both private sector and US government. These gaps have always been, to use a military-ism, wide enough to march a battalion through. The HR software was *supposed* to help narrow the gaps, which they identified as the reason for (costly) fit mismatches. Results? Mixed.

170

Reposted by Aaron Helton

evacide @evacide.bsky.social · 28/09/2026

Meta Muse appears to read your Apple messages and upload them to the cloud even if you explicitly tell it not to: appleinsider.com/articles/26/...

appleinsider.com

1001916986

Reposted by Aaron Helton

will jobst @willjobst.bsky.social · 28/09/2026

playing different games is good but playing with different people may be better

59516

Reposted by Aaron Helton

Thursday Garreau @keftiu.bsky.social · 28/09/2026

the OSR and PbtA scenes have some real common ground and i'm glad that some of the petty tribalism around both has died out. we get better games when folks take from multiple movements for their designs, imo!

1116223

Reposted by Aaron Helton

Tressie McMillan Cottom @tressiemcphd.bsky.social · 28/09/2026

I can’t say strongly enough that these are not realistic scenarios for the people I observed and, based on data about family lives, almost anyone else we might consider normal and/or not rich.

2986076

Reposted by Aaron Helton

Alicia @heyalicia.bsky.social · 28/09/2026

People handle tasks so complex and with so many shifting variables as part of our day-to-day existence that the complexity is underestimated. It’s like how knitting machines have existed for over 400 years but there still isn’t a machine that can crochet, though we think of those tasks as comparable

118928

Reposted by Aaron Helton

Earth & Environmental Sciences Ph.D Program-CUNY Graduate Center @eesgccuny.bsky.social · 28/09/2026

Curious about a PhD? Join our Earth & Environmental Sciences Open House on Thursday, October 8 in-person at the CUNY Grad Center or on Zoom! #CUNY #PhD #Admissions Details & registration: www.gc.cuny.edu/events/ees-a... @thegraduatecenter.bsky.social

Flyer for CUNY Earth & Environmental Sciences PhD Admissions Open House on Thursday, October 8 in-person at the CUNY Grad Center or on Zoom!

032

Reposted by Aaron Helton

Rachel Gunter, Ph.D. 🗃️ @phdrachel.bsky.social · 28/09/2026

I don't know how they think AI is going to take over the grocery shopping when most people aren't satisfied with another human doing the shopping on Door Dash/Instacart.

7616

Reposted by Aaron Helton

Dorothy Berry @dorothyjberry.bsky.social · 28/09/2026

An eternal bummer correlation to the necessary truth that there is no level of cultural milquetoast you can make yourself that will make you respectable enough to be heard in your profession, is that there is also no level of competency that will get you there either

091

Reposted by Aaron Helton

Albert Pinto @70sbachchan.bsky.social · 27/09/2026

@washingtonpost reported earlier that hospitals insurance costs next year are projected to spike by the most in 2 decades, partly "due to medical providers’ increased use of artificial intelligence to find more claims they can bill for" www.washingtonpost.com/business/202...

510966

Reposted by Aaron Helton

Bruno Dias @brunodias.dev · 27/09/2026

1. "Intelligence" in regards to living beings is at best a loose umbrella term for a diversity of behaviors and systems and at worst an eugenicist project 2. "Intelligence" in humans does not emerge strictly from symbolic manipulation, nor does it correlate with symbolic manipulation

3588

Reposted by Aaron Helton

Christina Wolbrecht @cwolbrecht.bsky.social · 20/09/2026

Delighted to see our Element, Gender & Local Political Ambition, out in the world this week! We ask whether local politics (with lower demands & closer to home) and appointed offices (which don’t require campaigning) are more attractive & accessible to women. 🧵 www.cambridge.org/core/element...

Cover image of the Cambridge Element, Gender & Local Politocal Ambition by C Wolbrecht, M Holman, L Iyer, S Chapa, F Rodger, & E Schroeder.

14312

Reposted by Aaron Helton

Hell Dunkin @halduncan.com · 27/09/2026

CSAMagotchi

03415

Reposted by Aaron Helton

Prisonculture @prisonculture.bsky.social · 26/09/2026

People sleep on CUNY but it is one of the absolute JEWELS of U.S. Higher Ed and has been for decades. And what's the thanks the system gets??? It has to fight for its life every single year because it has been systematically defunded.

23993181

Reposted by Aaron Helton

Gillian Branstetter @gbbranstetter.bsky.social · 27/09/2026

Reading a very strange and timely book from 1956 called The Obsolesence of The Human

410911

Reposted by Aaron Helton

Exhausted sociologist @iaracpassos.bsky.social · 26/09/2026

@mozilla.org launched a campaign against the tech billionaire class during the UNGA — and my face is in NY. 🧵 The ad places me next to Zuckerberg: he manufactures surveillance eyewear; I build local AI to expose political violence in Brazil, with Instituto Fogo Cruzado.

Triad of posters from the Mozilla Foundation campaign. First panel, left: Mark Zuckerberg; second panel, center: Iara, a white woman wearing glasses. Above, the text "Mark gets into eyewear / Gives creepers, peepers" over Zuckerberg and "Iara builds local AI / Exposes political violence" over Iara. Third panel, white background, all caps: "The powerful shouldn't choose what we see", below: "We all decide."

Triad of posters from the Mozilla Foundation campaign. First panel, left: Jeff Bezos; second panel, center: Rose Njeri Tunguru. Above, the text "Jeff builds a rocket. Celebrities go to outer space." over Jeff and "Rose builds democracy tools. Kenyan voices grow stronger." over Rose. Third panel, white background, all caps: "Joyrides won't build a better future", below: "We all will."

Triad of posters from the Mozilla Foundation campaign. First panel, left: Elon Musk; second panel, center: Rudy. Above, the text "Elon buys the blue bird. He trashes the place." over Elon and "Rudy builds Blacksky. Communities call the shots." over Rudy. Third panel, white background, all caps: "Wannabe kings don't rule the internet", below: "We all do."

35414

But can I take it to a job interview?

000

Damn, I didn't get that last job because I walked widdershins around the chair. Thankfully I counted. It was fewer than 7 times.

000

990000 @990000.bsky.social · 25/09/2026

Where's the lie? You know exactly how this would turn out.

010

Reposted by Aaron Helton

lnkd.in/p/gwq86W7w

Denis O.
Fintech Professional | Al Solution Ar...

The classical Al winter is here..

Everybody is running around with Jev or RLCD like it is the new kid on the block. Fast decisions, probabilities, confidence, structured outputs. No giant stream of generated tokens. Ooooh, ahhhh! Lol..
RLCD stands for Reinforcement Learning for Calibrated Decisions. But strip away the marketing and it is the familiar machine learning and statistics.. train a
model so its predicted probabilities are calibrated
against observed outcomes. But probability calibration is not a new concept. Neither are classifiers, scorers, discriminative models, confidence estimation, structured prediction or decision functions.
Yet somehow Al changed generations and forgot its own field.

Of course a system designed to make bounded decisions can outperform an LLM at bounded decision making. LLMs were never designed to be
stable autonomous decision engines in the first place. Go back to Attention Is All You Need. It was not a paper about AGI, autonomous intelligence or a
universal cognitive architecture. The Transformer was introduced as a sequence transduction architecture and demonstrated on machine translation. Then we became hypnotized by fluency at scale.. Hyperscale it enough, feed it enough
human output, and the statistical continuation over bullshit became so convincing that an entire industry started confusing linguistic competence with intelligence.. Then we wrapped it in tools, memory, RAG, planning loops and agent harnesses
and tried to turn a probabilistic sequence generator into an autonomous cognitive system.
And the industry response to Jev is hilarious. We are rediscovering classification, scoring, regression,
state estimation, structured prediction and bounded probabilistic decision making all over again when we already had all of this for decades if not a century (.!).

Jev may be useful engineering. Yet everybody is acting shocked that direct decision computation can be faster and more stable than asking a giant autoregressive LLM to generate tokens about the decision first.
It feels like the Al industry is suffering from dementia.. It went through machine learning, neural networks, reinforcement learning, probabilistic models, expert systems, cognitive architectures, search and planning.. Transformers came around,
scale exploded, and somehow everybody forgot where we had already been.. Now we are walking
around the same tracks in circle and calling it major progress.
The deeper problem is simple.

Hyperscaling did not breach the intelligence wall. It gave us extraordinary generative NLP, structured coding and generally better learned distributions through curated human labeled datasets.. But scaling the NLP did not magically turn it into a
fundamentally different form of intelligence or superduper AGI.. Stable autonomy under uncertainty
is still out of reach for GenAl.
So now the industry is moving backward while calling it 'forward' or Jev.. and that, my bros, is your
standard classical Al winter.

011

It's been a dream for the web since 1998, and it's been largely elusive since. And given the way things are going right now, I'm not sure it will ever be viable. Which means we'll be stuck instead with what some bloated platforms can summon statistically from the glut of info they've devoured.

010

Anyway, what RDF represents is a durable representation of asserted relationships that, within a framework of trust, can therefore translate to a durable representation of facts, distributed in a way that allows surprising questions but arbitrarily reproducible answers.

Apropos of nothing, I'm wondering why I never tried running Bakto's Terrifying Cuisine with the Wendy's TTRPG Feast of Legends. The PCs are made of food. (Which from certain perspectives was always true no matter the system...)

120

But the difference is that LLMs *seem* to be continually re-ingesting entire websites at the host's expense (one of several reasons the web is borked at any given time), especially if the data is not static. At the very least, distributed queries across RDF stores can (or should be) more targeted...

We can tell they're hammering websites because we can see the logs. There's otherwise no good reason for them to go fetch old data again and again to attempt to answer a question, except that they're (again, expensively) recompiling data they've sourced, ad infinitum.

Like other open systems, big players are happy to *consume* this kind of data. They just don't always contribute to the ecosystem of it, which is a consequence of their drive to centralize everything they can. Perhaps weirdly, LLMs attempt (at high cost) something similar to a semantic web query.

In its purest form, the semantic web is a massive web-scale database that you can leverage to run queries and joins and all that jazz, but because each thing you join with might be a different server, all those nodes make it weirdly fragile in ways that distributed systems are designed to mitigate.

From professional experience, I can tell you that machines *love* RDF, at least if my site stats are any indication. This stuff is made for machine traversal. But they do create an often unsustainable load on the system I have serving semantic web data, and they aren't slowing down.

When it's all done on the web, it's called the semantic web. The major downside to the semantic web as envisioned is that it requires a kind of infrastructure that bigger companies seem unwilling to make available and smaller players find too expensive to get up and running.

120

Moreover, it will give you that same data each time you ask for it with the same query unless the data itself has changed (e.g., added/removed/updated triples). Which means you could connect a natural language question to a means of generating this query.

This is most evident when the training data is sparse in the selected domain, because the models are generally tuned to say something rather than nothing. Querying a triplestore gives you what it can or nothing at all, or the closest match if you'v left it open to do so.

As the theory goes, armed with a set of trusted RDF triplestores, one can construct a query that will traverse the triplestores in precisely the same way each time. This is not how LLMs operate, because if you ask the same question in successive new sessions, LLMs give different answers each time.

Now, anyone can assert anything in RDF. Mistaken or even malicious assertions might find their way in. But an idea behind creating this web of interlinked assertions is that one might choose to trust or distrust a supplier of such assertions (RDF triples, or a graph).

But I started this with RDF. Because RDF is a means of encoding some kind of fact. Or rather, it's a means of asserting some kind of relationship between two things, with webs of interconnectedness and characterizations of those relationships. Semantic characterizations.

The data they're pulling from may be authoritative in some way, domain constrained, or whatever. But for the big models on the cloud, they are necessarily multi-domain. Which means while there certainly are facts within the data, they are only surfaced by probability and compute.