My latest: Thousands of people think their favorite chatbot has a soul, but why does no one ever say the same about Midjourney? Finally finished my essay debunking false beliefs in chatbot consciousness. Please share if you could: plus.flux.community/p/large-lang...

So clearly, what happened here was, they asked ChatGPT to crop the image, and since it can't actually do that if it's been fed the image as tokens instead of as a file, it generated a new image badly approximating the input. A great example of what happens when you let chatbots rot your brain.

Naomi Klein is always excellent.

Join us tomorrow!

ABAB

I am grumpy this evening

Do. Not. Use. Fucking. Sticker. Mule.

HAPPY FRIDAY TO EVERYONE WHO HATES SLOP

People keep saying “we can just fix hallucinated references”. They don’t understand that genAI hallucinates EVERYTHING: data, quotes, figures, you name it. References are just the easiest to catch. Friends don’t let friends write with genAI.

YOUR DAILY REMINDER PREVENTING POLLUTION PREVENTS HARM AND DEATH NONE OF THIS IS INEVITABLE www.theguardian.com/environment/...

Europe's worst heatwave so far, in terms of heat stress measures and temperatures - brought to you by fossil fuels. New @wwattribution.bsky.social study showing how such deadly heat became possible in only a few decades. www.worldweatherattribution.org/fossil-fuel-...

I spent like a day trying to figure out why my reimplementation of that METR chart doesn't have the exact same confidence intervals as theirs, and it turns out that the order that you feed the task runs into their bootstrapping method matters. Because of course it does. This methodology is so wacky.

'A narrative has taken hold that science and the humanities are at odds. In universities around the world, investment in cutting-edge technologies often comes at the expense of retrenchment in philosophy, history, literature and the arts.' 1/2

The AI-Tutor Revolution That Wasn’t www.theatlantic.com/ideas/2026/0...

The Republican president of the Utah State Senate lost his 2026 primary because of his support for "AI" Data Centers. UTAH. Democrats, i really really really need you to get this message and get it right now.

This is horrible.

Amazing. The bug in @leaflet.pub that prevents non-browser agents from accessing a publication with a custom domain, which I thought they fixed after I reported it to them in April, is still a thing.

METR's Time Horizon FAQ: "We launch 6 independent runs for each task" METR's actual results: 48% of tasks have fewer than 6 runs

METR only tested Time Horizon data against linear, exponential and upward hyperbolic fits, with an R² test. Since hyperbolic is clearly ridiculous unless the singularity is real and already happening, exponential is p much going to win by default unless the data points are in a very straight line.

Starmer saying the next PM will inherit a country that is 'fairer' than the one he inherited 2 years ago. In the one Starmer inherited, trans people could use public toilets.

TFW the example you use as a nightmare case of bad application of language technology has actually been implement (and in the works since 2019) in your own city. buttondown.com/maiht3k/arch... w/ @deccamuldowney.bsky.social

@timnitgebru.blacksky.app was really generous with her time when i asked her for this interview for @sciam.bsky.social #sciamyoungscientists. she has a lot of prescient things to say here. take a look. www.scientificamerican.com/article/timn...

tired of the "#1 dad" nonsense. if you're a dad, stop treating it like a joke and post your actual national dad ranking. i'm # 492,018, which is actually really high up there, but more importantly it lets me enter ranked dad lobbies where i can get matched up for fistfights with dads of my caliber

Making the notion of `computational intractability’ accessible is more important now than ever, because promises that AI systems can scale are false and misleading. This is not difficult to understand if one thinks soundly and from 1st principles. Paper: journals.sagepub.com/doi/10.1177/...

Andy Masley released a tool on his site to show how little of your total carbon footprint using a chatbot is. Except that it rates Claude models as having the lowest carbon footprint, which seemed strange to me given the compute bottleneck they're facing.

HEARTWARMING: This 11 year old developer still plagiarizes their games the old-fashioned way

If it seems like the ALA conference is heavy on boosters and light on critics, it's maybe bc they rejected all the critics

Our statement on the UK gov't's demand that all content on all devices sold or used in the UK be scanned, on the presumption of nudity, using a dystopian combination of age verification & content scanning. This proposal will not safeguard children. It endangers us all. signal.org/blog/pdfs/20...

I planned to finish my post on just how bad METR's Time Horizon data is a week ago, but the more I dig, the more I find. The original human baseliner results are in the runs.jsonl file on their eval-analysis-public repo, mixed in with the model runs.

Harder to think of a clearer example of "The Master's Tools Will Never Dismantle The Master's House" for this era than people using "generative AI" to protest the outputs and infrastructure of "AI"

“It's very obvious to me that the choices that are being made are going to throw many people into poverty [...] they won't be able to get out and go somewhere else, and people who are benefiting from this are going to move on." www.404media.co/a-farmer-don...

This is what happens when the people designing systems believe that chatbots can think. Now imagine this, applied to other domains where designers think the chatbot is a 'reasoning agent' that can be plugged into their workflow.

I can see this Lazy Taxi thing snowballing really hard and putting the fear into some execs Witty, insulting nicknames that roll off the tongue are marketing poison

So sick of purity tests. So he actively fantasizes about sexual violence in an incredibly disturbing way. So what? Who cares? Why do you hate the white working class? Have you seen his beard?

Everyone posting “It’s ok for Scorsese to use AI because it’s just for storyboards” is getting blocked immediately.

> I asked him whether he would rather stop writing than stop using AI in his writing process. “Yeah,” he answered. Just the concept that this is a mindset some people have makes me feel so tired.

I wish all users being charged for Microsoft AI a very You did this to yourselves. Suffer.

This is the Nvidia version of Microsoft’s Windows 95 launch. 😬

If you're anti AI & making claims that some of the models under the AI umbrella are ok because they seem to be serving scientific goals, then you fell for a type of hype. One from the 2010s; stats models aren't blessed by their use in science and they can erode our thinking. bsky.app/profile/oliv...

Another one. These people must be driven out of polite society before it is too late www.theguardian.com/books/2026/m...

Declaring oneself to be the GOAT is quintessentially cringe. You do not get to call yourself the GOAT. It is a title that must be bestowed upon you either by a broadly respected domain expert, or by overwhelming public sentiment.

Leftists don’t hate technology, actually. About that silly article...

there is something quite ironic about AI slop selling a NIMBY message

Extraordinary. Tim Dunlop admits to using AI to generate a story about Matt Chun and the Bondi shooting, with quotes attributed to Chun that he never said. Entirely fabricated. By AI. The levels of harm and stupidity are enormous. #AusPol open.substack.com/pub/tdunlop/...

Paramount+ wanted to use the image on the left as a poster for the wrath of Khan, so they used everyone's least favorite technology to generate the rest of the image.

AI is largely developed by men with uses that are biased against & predatory toward women. Faced with threats to their livelihoods, many admin assistants have felt forced to use it. Women make up 86% of those in jobs vulnerable to be replaced by [email protected] 19thnews.org/2026/05/wome...

Some confusion between copyright and trademark laws in the comments here! You can definitely parody a trademark in your creative work. You can't however register the parody as an overlapping trademark of its own, which is what Pattie Gonia is trying to do.

Note how at no point does the quoted individual actually point to a simple pattern like "it's not X, it's Y". This narrative that simple surface-level pattern matching is the only way that readers can tell the text did not come from a mind is a way to normalise extruded prose as human-equivalent.

Yikes www.hollywoodreporter.com/tv/tv-news/g...

Google AI overview: It's like a page from a deranged children's book