Monday June 16th, 2025
LLM users also struggled to accurately quote their own work. While LLMs offer immediate convenience, our findings highlight potential cognitive costs. Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels. These results raise concerns about the long-term educational implications of LLM reliance and underscore the need for deeper inquiry into AI's role in learning.
Via Gernot Wagner @gwagner@fediscience.org:
To write is to think. Using ChatGPT to write leads to..."cognitive debt", which might be one of the better euphemism for somewhat less polite words.
Small n, not yet peer-reviewed, etc https://arxiv.org/abs/2506.08872
Sunday June 15th, 2025
Just add humans: Oxford medical study underscores the missing link in chatbot testing
A paper by researchers at the University of Oxford found that while LLMs could correctly identify relevant conditions 94.9% of the time when directly presented with test scenarios, human participants using LLMs to diagnose the same scenarios identified the correct conditions less than 34.5% of the time.
Perhaps even more notably, patients using LLMs performed even worse than a control group that was merely instructed to diagnose themselves using “any methods they would typically employ at home.” The group left to their own devices was 76% more likely to identify the correct conditions than the group assisted by LLMs.
Because of my mom's various theories over the years, I'm kinda familiar with this, but it's good to have an overview. NPR: Ancient miasma theory may help explain Health Secretary Robert F. Kennedy Jr.'s vaccine moves, but Ars Technica: RFK Jr. rejects cornerstone of health science: Germ theory really gets more nuanced on miasma theory vs terrain theory.
I think the good news about this whole era is that people I know who were all-in on homeopathy are now going "wait a minute", and my own thinking has evolved from "maybe the raw milk I grew up on wasn't all that bad, it sure tasted better" to "yeah, not worth it."
Joshua Conkel @joshuaconkel.bsky.social
We should make June 14th No Kings Day permanently.
Thread on event planning and yesterday's Victory Day Trump Birthday parade
So yesterday, reports are that Oakland PD worked with the crowds in the protests to keep them civil, and the LAPD and LASD exchanged fire like the warring gangs they are, and blamed it on the protesters.
Guess Oakland learned something from The Riders that LA hasn't yet.
Wait, the attempt to replicate the USSR Victory Day parades sponsored by cryptocurrency? So like literally brought to us by North Korea?
I'm trying to do some stuff with steel scavenged from an old bed frame. It is absolutely tearing my drill bits up, but I got through, and then couldn't get the bolt in the hole. I remeasured, everything should fit, then I looked at the leading 3/8" of my drill bit, and ...
I'm trying to do some stuff with steel scavenged from an old bed frame. It is absolutely tearing my drill bits up, but I got through, and then couldn't get the bolt in the hole. I remeasured, everything should fit, then I looked at the leading 3/8" of my drill bit, and ...
Saturday June 14th, 2025
No Kings rally in downtown Petaluma, we've just gotten past E At, looks like it extends up along D St to at least 5th
Friday June 13th, 2025
Of course these days it's hard to tell what's real and what's a troll, but bwahahahahaha: Cursor Forums: Cursor YOLO deleted everything in my computer
(From that thread, how to fuck up your repo with VSCode)
With lots of conversations recently about LLMs, and how people don't seem to have the tools to process what they're seeing, I'm pondering a lot about Baldur Bjarnason's note that avoiding generative models is the rational and responsible thing to do which is a shorter follow-up to Trusting your own judgement on ‘AI’ is a huge risk as I hear a lot of conversations by smart people who, if pressed, would say "these things are a clever hack", but then go on to use them as though they're omniscient entities.
Referencing a Reddit conversations about people losing perspective with ChatGPT, Elf Sternberg highlighted:
"Being able to say 'That sounds nuts' without having a point by point rebuttal is a critical talent for surviving in this world." AIS are not meant to survive in this world.
and as the poster of those conversations Linnea Sterte @decassette.bsky.social noted
some of these are prob mostly lies ppl made up but the 'no I'M the one who made the ai awaken & become sentient' via eliza effect or w e is fascinating as a weird cyberpunk narrativesome of these are prob mostly lies ppl made up but the 'no I'M the one who made the ai awaken & become sentient' via eliza effect or w e is fascinating as a weird cyberpunk narrative
Even the NYT is starting to notice: Justin Hendrix @justinhendrix.bsky.social
Here is a gift link to @kashhill.bsky.social's must read piece on the dangers of OpenAI's sycophantic LLMs and the parasocial relationships people are creating with them.
New York Times: They Asked an A.I. Chatbot Questions. The Answers Sent Them Spiraling. (Gift link)
“What does a human slowly going insane look like to a corporation?” Mr. Yudkowsky asked in an interview. “It looks like an additional monthly user.”
Nikki McCann Ramírez @nikkimcr.bsky.social
Padilla: If this is how DHS responds to a senator with a question you can only imagine what they're doing to farm workers, to cooks, to day laborers out in the Los Angeles community and throughout California.
(With video)
Fuck ICE: Over 200 arrested in L.A.; National Guard deployed in Texas as rallies continue nationwide
The confrontation occurred after officials from Downey Memorial Christian Church and others confronted a group of five armed men in plainclothes and tactical gear who “swarmed” a man sitting under a tree in the church’s parking lot, Al Lopez, a pastor at Downey Memorial, told reporters.
Lopez said he asked the men to identify themselves, but they refused.
“They kept asking us to step back and telling us that we couldn’t be there,” he said. “When we said, ‘We don’t want this on our property,’ this gentleman just shouted again: ‘The whole country is our property.’”
Thursday June 12th, 2025
CHOAM Nomsky @samthielman.com
wow I wonder if the cops who lied about the sitting senator even though everything they did was on tape are also being less than honest about the presence of violent agitators at the protests in los angeles. they might even not have told the entire truth about other protests
Dear presentation software makers: *Please*, for the love of all that's holy, give me a mode where I can see the slide full page *without* going full screen.
Signed: dude that's trying to do Zoom broadcasts with slides and really hates that I don't have OBS controls when that's happening.
The clusterfuck invasion of California as of right now:
Guard Soldiers Deployed in Trump's LA Crackdown Aren't Getting Paid Yet. (Via
Chairman of the Joint Chiefs contradicts Trump’s takes on Putin, LA protests (Via)
Justin Baragona @justinbaragona.bsky.social
Kristi Noem: "We are not going away. We are staying here to liberate the city from the socialists and the burdensome leadership that this governor and that this mayor have placed on this country and what they have tried to insert into the city."
Sen. Alex Padilla is then forcibly removed!
(With video)
Video of the assault on Padilla.
holy shit -- GOP Rep. Mike Turner tries to "help out" Hegseth by getting him to confirm that the Pentagon does not in fact have plans to take Greenland by force, but Hegseth instead more or less testifies that yep, they have devised plans for that
"The Cloud" was a mistake.
(This message brought to you by Firebase being down.)
Is this butter "cultured", Or just pretentious?
Deciphering Glyph: I think I'm done thinking about genAI for now
...I cannot effectively respond to these folks, because they are making a practical argument that I cannot, despite my best efforts, find compelling evidence to refute categorically. My experiences of genAI are all extremely bad, but that is barely even anecdata. Their experiences are neutral-to-positive. Little scientific data exists. How to resolve this?
Gabbard says AI is speeding up intel work, including the release of the JFK assassination files
“We have been able to do that through the use of AI tools far more quickly than what was done previously — which was to have humans go through and look at every single one of these pages,” Gabbard said.
I really really hope that these were LLMs hosted on a secure machine, but even so...
Wednesday June 11th, 2025
Closing off tabs: Just make it scale: An Aurora DSQL story
At re:Invent we announced Aurora DSQL, and since then I’ve had many conversations with builders about what this means for database engineering. What’s particularly interesting isn’t just the technology itself, but the journey that got us here. I’ve been wanting to dive deeper into this story, to share not just the what, but the how and why behind DSQL’s development. Then, a few weeks ago, at our internal developer conference — DevCon — I watched a talk from two of our senior principal engineers (PEs) on building DSQL (a project that started 100% in JVM and finished 100% Rust). After the presentation, I asked Niko Matsakis and Marc Bowes if they’d be willing to work with me to turn their insights into a deeper exploration of DSQL’s development. They not only agreed, but offered to help explain some of the more technically complex parts of the story.
Because I'm going to want to find this later: A Letter from the Episcopal Bishops in the State of California
Like all Californians, we are watching with great concern the events unfolding around immigration protests in Los Angeles. We are deeply concerned about the ICE raids and about the potential for violence arising from the deployment of National Guard troops and Marines to the Los Angeles area. We are concerned that military deployments will escalate the confrontations unnecessarily, and worry that all of our regions in California may be subject to future deployments that heighten tensions rather than resolving them.
ChatGPT "Absolutely Wrecked" at Chess by Atari 2600 Console From 1977
In a post on LinkedIn, Citrix software engineer Robert Caruso explained how the OpenAI chatbot "got absolutely wrecked" by an Atari 2600 running Atari Chess, a game for the system released in 1979, when Jimmy Carter was still president.
Disclosure: Covert Web-to-App Tracking via Localhost on Android. If you're familiar with peer discovery protocols....
On or around May 17th, Meta Pixel added a new method to their script that sends the _fbp cookie using WebRTC TURN instead of STUN. The new TURN method avoids SDP Munging, which Chrome developers publicly announced to disable following our disclosure. As of June 2, 2025, we have not observed the Facebook or Instagram applications actively listening on these new ports.
“Localhost tracking” explained. It could cost Meta 32 billion. (Let's fucking hope.)
Just to log such things here: I've been getting all of the "use LLMs" pressure, and work has me using the OpenAI API. I've got a couple of features I want to add and clean up in my C++ static site generator, and thinking maybe I need to rework that into Rust, in order to learn Rust, so along with my explorations in parsers in Rust, I asked gpt-4o:
I need command line application written in Rust. It should take as arguments one or more paths to search. It stores all of its information in an SQLite database, and if the SQLite database doesn't exist, it can create the database.
It should recursively search the specified paths for files. If the last modified date on a file is more recent than the entry in the SQLite database, or does not exist in the SQLite database, it should calculate a SHA512 sum of each file. If those files are a well known image format, it should extract the height and width of the image. It should store filename, the filename with full path, last modified date, checksum, height and width of each file in the database.
And I got... something that looked like a Rust application, as Cargo.toml and main.rs, but was not in fact a Rust application. It did, however, seem like a reasonable bit of code to fix, so after adding a bunch of <()> to -> Result return specs, and doing a lot of unwrapping and match and error return type management, I've got something that ... runs.
I also had to change how it did file date checking, because it got the semantics of SystemTime totally wrong.
cargo build pulls in a huge list of modules that I don't know how to trust or evaluate (dependencies of dependencies, unlike my previous attempts at vibe coding Rust this did seem to only suggest semi-relevant modules), it seems to run slower than forking jhead and pnginfo processes to find image sizes in C++ (my guess is that the image::GenericImageView is loading the full raster, and doesn't need to).
So... interesting, but not sure how helpful it is vs reading the docs yet.
I'm in email correspondence with someone who's suggested that I should automate older links on Flutterby tying to the Wayback Machine. And it's just a little bit of JavaScript, could be tied to a checkbox or something at the top of the page and a cookie, but it's also got me thinking about how much an individual site attempts to override a user's browsing behavior.
One of the things that drives me fucking nuts about software is when the software thinks it can do better than the plaform services, and creates additional layers on top of the system in ways that attempt to replace (and often interfere) with the underlying systems.
And it could go further, I'll sometimes link to archive.ph or similar services if something has a paywall, which I also sometimes feel kinda guilty about (although it seems like it's generally "a take"/dunking on the article, so...), is it reasonable to do that?
Or do I just assume the sophistication of users to derive such links through whatever means they have (manually, plug-ins, whatever)?
I'm just here for the headline: Financial Times: Duolingo CEO on going AI-first: ‘I did not expect the blowback’. The rest of the article is a puff piece, archive.ph version.
Hmmm.. this seems like the sort of thing I'd have linked, but I don't see it. From 2020: Journal of Transport & Health: Estimated car cost as a predictor of driver yielding behaviors for pedestrians
Results Of 461 cars, 27.98% yielded to pedestrians. Cars yielded more frequently for females (31.33%) and whites (31.17%) compared to males (24.06%) and non-whites (24.78%). Cost of car was a significant predictor of driver yielding (OR = 0.97; p = 0.0307); odds of yielding decreased 3% per $1000 increase.
Via.
Prof. Sam Lawler @sundogplanets@mastodon.social
The more I learn about atmospheric chemistry, the more terrified and angry I am about satellite companies' blatant lack of consideration for how their actions will harm the atmosphere. I hope this gets a lot of press. Great work by a whole team of scientists, including @astrokiwi.bsky.social!
npj Climate and Atmospheric Science: Near-future rocket launches could slow ozone recovery.
Ryan Castellucci @ryanc broke github with a shell injection in the committer email message and bad timestamp in a git commit. Other people in the thread are looking at it as a possible way to kill the repos of people who fork a project and then SEO their fork above the original...
myrmepropagandist @futurebird@sauropods.win
This whole AI thing is like being in a restaurant with a fancy waiter who has one of those giant pepper shakers, only it's full of bird poop and anthrax and he keeps asking "would miss like a little AI on her pasta?" with every dish that comes out... and when you say "NO" he starts grinding anyway and won't stop until you physically knock him over.
Hmm, parking in downtown LA *is* really bad, the Marines had to park in Seal Beach (Orange County). If they want a better route to the Civic Center, I'd hop on Amtrak and hop off at LA Union Station. (it's a lot better).
Marines ordered by Trump to serve in Los Angeles arrive in ... Orange County?
Thomas Sturm @tsturm@famichiki.jp
@ai6yr "LA cant be occupied by armed forces as there is no parking." is the most California take possible.
😂
Tuesday June 10th, 2025
Gonna name a git branch my_finger so I can tell people to pull it.
As Apple's WWDC keynote this year highlights... uh... a new "design language" that has contrast issues and is gonna result in the Apple users in my life asking me where the fuck the buttons they expected to find have gone, and suddenly it's my problem to figure all that shit out. And all sorts of other features that nobody actually cares about...
A good rundown on the ways that big tech computing has fucked up design: Adrian Roselli: I Don’t Care What Google or Apple or Whoever Did.
Wall Street Journal: The Pentagon Disinformation That Fueled America’s UFO Mythology
U.S. military fabricated evidence of alien technology and allowed rumors to fester to cover up real secret-weapons programs
Summarized by Valerie Aurora 🇺🇦 @vaurora@mstdn.social
Surprise. All the US military UFO stuff turns out to be deliberate misinformation to cover up advanced military projects, or shitty Air Force hazing rituals
Sources of data that haven’t been contaminated by AI-created content. Low Background Steel (and lead) is a type of metal uncontaminated by radioactive isotopes from nuclear testing. That steel and lead is usually recovered from ships that sunk before the Trinity Test in 1945. This blog is about uncontaminated content that I'm terming "Low Background Steel". The idea is to point to sources of text, images and video that were created prior to the explosion of AI-generated content that occurred in 2022.
The Register: Even modest makeup can thwart facial recognition
In a pre-print paper titled "Novel AI Camera Camouflage: Face Cloaking Without Full Disguise," David Noever, [PeopleTec] chief scientist, and Forrest McKee, data scientist, describe their efforts to baffle face recognition systems through the minimal application of makeup and manipulation of image files.
Futurism: People Are Becoming Obsessed with ChatGPT and Spiraling Into Severe Delusions. Got a good rundown of other resources with similar results.
In other words, OpenAI has all the resources it needs to have identified and nullified the issue long ago.
Why hasn't it? One explanation echoes the way that social media companies have often been criticized for using "dark patterns" to trap users on their services. In the red-hot race to dominate the nascent AI industry, companies like OpenAI are incentivized by two core metrics: user count and engagement. Through that lens, people compulsively messaging ChatGPT as they plunge into a mental health crisis aren't a problem — instead, in many ways, they represent the perfect customer.
Monday June 9th, 2025
Wilhelm Fitzpatrick @rafial@masto.hackers.townM>a<
Sitting here with gasping like a fish after reading a detailed description of using an AI coding tool to write a unit test for a simple function in which it is mentioned:
* The AI was apparently unable to "reason" about correct dependency injection or mocking
* 20 prompts were written to "iteratively refine" the solution
* 2000 lines of code were generated of which 50 were used...and at the end the writer concludes "the final result was successful and saved time over doing it manually"
😱
Wilhelm Fitzpatrick @rafial@masto.hackers.town
I continue to be astounded by how LLMs seem to hack the basic reasoning processes in our own meat brains!
Wow. I know I'm disappointed and disgusted with the tech industry lately, but the #wwdc25 examples are completely falling flat with me. "AI can help you shop better!" "You can put backgrounds on your text chats!"
If this was on The Onion, I'd understand, but while MacOS has so many bugs...
So glad there's only one iOS user who I do tech support for in my life. The "where'd my button to do this go?" questions are gonna be numerous.
I feel this so hard. Especially as the Mac just gets worse with every release: afreytes 🇵🇷 ☭ @afreytes@mastodon.gamedev.place
I'm sorry in advance if this sounds cringe or sappy. Or if it is something well known. But right now the difference, for me, in using Linux versus Windows is hope. It really is. Let me explain.
On Linux, if I have an issue, or a problem, something I don't understand. I have hope that I can find a way, an alternative, a forum, someone that will help, or even make it better by myself. No matter the issue hope drives me forward.
On Windows, there is no hope anything will get better.
Media changes definition of ‘crossfire’ to include when a cop points a gun at you and shoots you
This comes after Australian journalist Lauren Tomasi was shot by police with a rubber bullet while covering the protests in LA.
The incident which was caught on camera and shows the officer look at the reporter, then pointing a gun directly at her before shooting her, has been described by outlets including her employer Channel 9 as being caught in the ‘crossfire’.
Whoah. Checking the Petaluma forecast for tomorrow, NWS says high of 77, Weather Underground says 70, AccuWeather says 76, Weather Channel says 69. That's quite a spread.
Simon Tatham @simontatham@hachyderm.io
Who called it "cloud identity provider" when they could have called it "Attack Surface as a Service"
MostlyHarmless @MostlyHarmless@thecanadian.social
When you stare into the abyss and the abyss, without a word, hands you your usual order right away.
Mark Carrigan @markcarrigan.net@markcarrigan.net — A depressing fable about how ChatGPT is corroding trust in scholarship. In which he discovers that the opening quote of Trust: Building on the Cultural Commons by Pascal Gielen inadvertently(?) starts out with a likely LLM fabricated quote attributed to bell hooks, but probably isn't.
OpenAI is retaining all ChatGPT logs “indefinitely.” Here’s who’s affected.
In a statement, OpenAI Chief Operating Officer Brad Lightcap explained that the court order came in a lawsuit with The New York Times and other news organizations, which alleged that deleted chats may contain evidence of users prompting ChatGPT to generate copyrighted news articles.
So, uh, yeah, careful there kids, your chat logs are fair game for law enforcement, no matter what you think you've told OpenAI to do with them re retention.
Buncha klan members got in hopped up trucks, covered their faces, and went down to the Home Depot to kidnap and harass brown people. Motherfuckers should be glad the community let them get away with their lives.
The responsibility for violence lies with the instigators. They need to be told this.
Wandered around the Martin Civic Center lagoon, there were turtles and birds heron there...
Sunday June 8th, 2025
Yesterday my BlueSky and Fediverse feeds lit up with reports of an ICE vehicle driven at high speed into a crowd. Today I can't find news reports on this event. Anyone got details?
When the LAPD thanks the protest organizers. We need to get ICE the fuck out of California.
nova (they/them) @nova@sharkey.stellar.gay
correlating a lack of critical thinking in 18 year olds with chatgpt instead of the holistic failure of the education system is such a joke and insult to kids, honestly
the failure didn’t start when they were 15 and chatgpt became commonplace
the failure started at the start of their education, when the sole intention of the system is to produce malleable workers who do not question authority
we are failing our kids and have been since long before chatgpt. stop acting like it’s not the fucking design of the education system to make a complacent public.
The national emissions estimate they arrived at—230 kilotonnes per year—is sevenfold higher than the 34 kilotonnes reported in Canada's National Inventory Report. The study was published in Environmental Science & Technology.
Of particular interest to me 'cause I know someone caught up in the drama of an Alberta well...
Effin' Birds: Statement on Unbound
John Mitchison, Unbound’s publisher, posted a heartfelt apology yesterday, announcing that he was stepping away from the business because authors are not getting their money. I feel a lot of empathy for John, because he, too, lost his dream to Unbound’s financial mismanagement. But the act of destroying a creative person’s confidence to hide the company’s monetary losses is unforgivable.
It took me two years to get over the feeling of rejection that came from a book that, it turned out, was successful and should have had a second printing.
Through extensive experimentation across diverse puzzles, we show that frontier LRMs face a complete accuracy collapse beyond certain complexities. Moreover, they exhibit a counter-intuitive scaling limit: their reasoning effort increases with problem complexity up to a point, then declines despite having an adequate token budget.
🏳️⚧️🏳️⚧️🏳️⚧️*looks at you 俺ly*🏳️⚧️🏳️⚧️🏳️⚧️ @kirakira@furry.engineer
*invalidates your cache*
Here we used a large international sample (N = 2035) and novel within-participants testing to show, for the first time, at least two environmental pathways linked to judgement biases: one related to people’s social surroundings and linked with their explicit views on transport, and a separate, more implicit pathway related to higher-level structural influences such as nationality, and living in rural areas. Additionally, respondents dramatically underestimated public support for non-motorised transport relative to their own, a pluralistic ignorance effect likely reflecting another facet of motonormativity. The social-ecological explanation, with its nested environmental influences, helps explain the ‘stickiness’ of automobility, and implies change will be most likely when multiple facets of a person’s social, physical and cultural surroundings align in supporting non-motorised mobility.
Adrianna Tan @skinnylatte@hachyderm.io
If you missed my ‘From FinTech to Fin Tech’ talk at North Bay Python, here’s a blog post with the video
The slide everyone was talking about is at 06:27
Saturday June 7th, 2025
I remember when "web scale" was low single digit millions of queries per day. Now that's table stakes for a blog site to keep up with the crawlers, even as search has gone to complete suck.
20mph limits in London linked to sharp fall in road injuries and deaths, new report finds
A new study published by Transport for London (TfL) has shown that the introduction of 20mph speed limits and zones on local authority-managed roads in London between 1989 and 2013 led to significant reductions in collisions, injuries, and deaths, particularly among children and vulnerable road users.
Google AI definitely has a notion for the lore and mythology of elves in Dungeons and Dragons.
Anti-Petaluma Appellation Hotel advocates: "I'm not against a hotel in principle, just that tall."
Also: "We can't cover up the mural on the side of the hardware store!"
So apparently the appropriate height for the hotel is zero feet.
"Circle garden is a temporary project that transforms an existing traffic-slowing element into an appealing green space. ..."
Wide shot of the 5th St traffic circle garden.
Great work 5th St folks with the garden in the traffic circle at G St.