Tokenmaxxers getting meatmogged by brain-only writers whenever us-east-1 goes down

Almost every week I think about the end of my podcast with @aidanwalker.bsky.social about where this all ends... www.theatlantic.com/podcasts/202...

AI-writing tropes make reading LLM synthesized text so freaking monotonous

Hallucinated references seem to be way more common in social science papers than others - but why? My latest for @nature.com www.nature.com/articles/d41...

when people ask me how I think AI is going to go I always start with "well, it's the most powerful surveillance technology ever invented..."

This headline is underselling that this is a form of instrumentalized surveillance and behavioral control (which is the social and historical design context of “AI” technologies)—not merely predicting personal individual behaviors but recommending them through suggestion.

Yeah, well what if we had an artifact that wasn’t entangled in its social and historical context? Checkmate, haters

Just like me fr fr

I’m still mad the librarians haven’t made it a whole weird thing why are there not more of us on here

Backlash as Utah approves datacenter twice the size of Manhattan. Facility would require more power than entire state uses and suck up vast amount of water in drought-stricken area. www.theguardian.com/us-news/2026...

NEW: More than half the U.S. is in drought. But few places are dealing with the kind of crisis facing Corpus Christi, Tex., which could see water demand outstrip supply w/in a year. One official told me: "This is actually the canary in the coal mine. 🎁: www.nytimes.com/2026/05/13/u...

A super important - and really troubling - thread by @caitlindeangelis.bsky.social revealing that FamilySearch's AI transcription is deracinating and sanitizing references to race and enslavement in historical records. Every historian and genealogist should read this 🗃️

I don't usually use Chrome, but I had to for a reason, and it is remarkable that between the Chrome browser itself and the Google search home page, they now have four different buttons coercing the user into using AI all on the screen at once.

Okay. I realized I was posting too much today, so now this exists as a thing: stechometer.neutral.zone

My take on this as a person who spent their 20s in the same headspace (and was actually one of RAW's students during the brief time he was teaching classes) is all the "reality is what you can get away with" stuff is exactly correct—and if pursued uncritically gets us exactly what we got.

A little-noticed X account in India was struggling to reach 100 followers when it announced: "Sit tight and have a view while I am working my way up the ladder." Two years later, it's the account Elon Musk interacts with the most. It lavishes praise on Musk and spreads his ideas. wapo.st/4nqjRYk

Nearly 50,000 people in the Lake Tahoe area have been told that their utility will stop providing power to them, because it's redirecting that power to data centers. /1

15/ Substantive takeaway: LLMs separate the message from the messenger. State-coordinated phrasing can circulate through the web, enter training data, and reappear as neutral-sounding LLM output— the source obscured.

New paper in Nature. The more a government controls its domestic media, the more it dominates AI training data, the more pro-regime outputs we get from AI. By scraping the open web, LLMs are unwittingly laundering state-coordinated narratives into seemingly objective answers.

the thing these all have in common is that until recently you had to be a loser to know anything about them. and the nexus of that loserdom is rationalists

This is the secret to automation that bosses refuse to understand: There are no* "tasks that are easy for people" and "tasks that are easy for AI." The tasks that AI fails on are the same tasks that are difficult for a person - except now they don't have context or the expertise to do them.

OK, everyone just writes their own software and generates their own media and companions, and reposts slop, and sends their little agents out to fight with other agents, total solipsism with no financial incentive to do anything new of quality for public consumption, what's your value prop?

"Predictably, the huge spike in productivity that these companies claim their own AI products have enabled hasn’t resulted in more or better products, shorter work weeks, or better consumer experiences." www.404media.co/software-dev...

The real tell is that no one is talking about using "AI" to make *better* products/services/media for customers. There's real "you'll get nothing!" energy coming out of these organizations. "Everyone can just make their own stuff" is a flavor of you'll get nothing.

Turning on this spigot made everything wet! Maybe if I turn it the other way it can make everything dry?

I think it’s hard to square the fact that social media (and the digital new media) has been both indispensable for the rise of MAGA and emergence of conspiracism as a new political style and force, etc, while simultaneously being incapable of driving other political changes, like constraining power

Post-information society

I think the problem with posting, and we've got ample evidence of this now going on 20ish years, is that it's apparently a lot easier to make people evil and crazy with posts than it is to make them decent and kind.

Extremely funny object to me. Brb hanging up my framed poster that says “I like oil paintings”

Also now available as a $20 reproduction on maple wood veneer is Prodigium Regis Flavi et Exitii Carcosae -- Sign of the King in Yellow and the destruction of Carcosa. 8.5x11" so it will fit standard frames of that size. #art #bsnm xylographilia.etsy.com/listing/4310...

The Diamond Age, or The Young Ladies’ Illustrated Predator

The DOGE ideology continues unabated

oh good, we're letting the AI choose who gets to eat

Gonna log off for another few weeks maybe

NEW: The NYT sent a firm email to freelancers today warning that "contributors must not submit any material for publication that contains content generated, modified or enhanced" by generative AI. The "reminder" notably follows a string of AI incidents at the paper: futurism.com/artificial-i...

If you’re into books, may I suggest Bookhive, a Good Reads-like reading tracker that is easy to use, open source, runs on AT Proto and is developed by very nice people? I’ve been enjoying using it. bookhive.buzz

Putting solar panels in orbit, or even just talking about it, is a way to distract from putting solar panels over parking lots, over canals, on roof tops. It's doing some kind of political work, and that work is at least adjacent to fossil fuel interests.

There is no way to do anything interesting in any sustained way in any field without immersion in that field. That's just a fundamental fact of being human, and there's no getting around it.

that the inability of models to parse sarcasm isn't just a misinformation risk, it's a *dis*information risk. Bad guys have been fucking with academics and cops for decades by mixing memes and what they actually believe. Tricking AI is child's play in comparison.

You absolutely cannot ask chatGPT what it means, because chatGPT will string together a series of words that sounds plausible, and deliver the answer with confidence, and it will be completely wrong.

While I’m pretty sure this article was written by AI, this is a good example of the specific way that LLMs can’t actually analyze data, and, consequently, aren’t actually summarizing text (which is not an act of synthesizing statistical word bags but grounding truth claims in evidence).

I often rely on LLM summaries of meetings and other events or materials. But my eyes were opened by this insightful, stern critique of the methodology by which LLMs “summarize”: William Gleng, “LLM Summarizers Skip the Identification Step”: towardsdatascience.com/llm-summariz...

Somewhat related, but when I got a B.S. in computer science 20+ years ago my advisor, who was the head of the department, threatened to kick me out of the program because I wanted to do an independent study course in AI, and the only reason he didn't go through with it was because he retired first.

This winter I taught a new course entitled "Politics of Digital Technologies" and we had tons of fun discussing a range of digital policy issues. I'm making the syllabus & reading list open access, hope it's helpful for your own teaching or learning! Materials here: www.blairaf.com/resources/

knightcolumbia.org/content/ai-a... Now up at @knightcolumbia.org - Cosma Shalizi and myself on "AI as Social Technology." It says that AI models are really a technology of social mediation that allow humans to connect with thoughts of other humans. Short-ish thread with major points follows.

slightly increase a band Aphex Triplet

a primal scream that captures a moment