Sunday April 20th, 2025
Happy Bicycle Day and 4/20 Eve, y'all!
Fun day with the quadricycle at the Petaluma Butter and Egg Days parade, in it with the Safe Streets Coalition folks, and then hanging out with Sean and friends at LivXplore.
Saturday April 19th, 2025
OpenAI’s new reasoning AI models hallucinate more.
In its technical report for o3 and o4-mini, OpenAI writes that “more research is needed” to understand why hallucinations are getting worse as it scales up reasoning models. O3 and o4-mini perform better in some areas, including tasks related to coding and math. But because they “make more claims overall,” they’re often led to make “more accurate claims as well as more inaccurate/hallucinated claims,” per the report.
It's interesting that we're using terms like "reasoning" in conjunction with machines "hallucinating". Like, when I see a person on the street ranting at the sky I am not thinking of their behavior as connected to "reasoning".
A careful read of this article is also demonstrating all of the ways in which OpenAI has managed to define success for itself...
Friday April 18th, 2025
Docks are filling up for the parade tomorrow...
Interesting to watch Slack social groups migrating to Signal...
As ‘Bot’ Students Continue to Flood In, Community Colleges Struggle to Respond
The bots’ goal is to bilk state and federal financial aid money by enrolling in classes, and remaining enrolled in them, long enough for aid disbursements to go out. They often accomplish this by submitting AI-generated work. And because community colleges accept all applicants, they’ve been almost exclusively impacted by the fraud.
Via.
As far as I'm concerned, the only legal definition of a woman should be:
When Shania Twain says, "let's go girls", do you go? If so, girl.
Kelly McBride at NPR: How does NPR cover peaceful protests when the only news is the protest?
"Not Very Compelling": How NPR Dismissed the Largest Protests of 2025
McBride's position essentially argues that mass protests only become newsworthy when they turn violent or disruptive. She writes that “once a protest movement results in conflict or property damage, NPR journalists covering the protests will often note the exception.” This creates a perverse incentive: want coverage? Create conflict.
I am a banana @[email protected]
You telling me these costs are denser than 1g/cm^3?
Large Heydon Collider @[email protected]
"This AI output is highly inaccurate."
"Nah, you're just prompting it wrong."
"How do I go about prompting it the right way?"
"You really need to know the subject you're asking about. Then you can help it avoid making mistakes."
"If I know the subject deeply myself, why am I asking an AI about it?"
"It helps to train the AI."
Private chat with friends who work in public health talking about Trump initiatives in the FDA and CDC, and how there's lots of indication that nobody in the current administration understands how things already happen there.
And I'm seeing lots of parallels between that and the Nextdoor mobs.
Company apologizes after AI support agent invents policy that causes user uproar:
On Monday, a developer using the popular AI-powered code editor Cursor noticed something strange: Switching between machines instantly logged them out, breaking a common workflow for programmers who use multiple devices. When the user contacted Cursor support, an agent named "Sam" told them it was expected behavior under a new policy. But no such policy existed, and Sam was a bot. The AI model made the policy up, sparking a wave of complaints and cancellation threats documented on Hacker News and Reddit.
Seems like maybe this started with an over-zealous anti-fraud or security measure, but LLM based support agent misrepresenting this blew it up. As Reddit user BrokenToasterOven says:
I literally just cancelled my sub and moved to AUGMENT CODE.
I was dumping like $700/wk into Cursor through work, and now we're purging it completely.
Which, ya know, maybe that user walks it back, maybe they don't, maybe they've already found the other product that they like, but this is a liability...
They are all very nerdy. So they deserve a spot on my nerd blog. But they are also very arty, so they deserve to be on my Love Nonsense blog as well. I chose to write about my clocks on Love Nonsense, so here’s a summary of all the clock posts I wrote over there.
Stevens: a hackable AI assistant using a single SQLite table and a handful of cron jobs.
The LLM bits of it seem superfluous, but it's interesting to see how people are playing with this stuff. There was a discussion on Facebook yesterday about making a thing that you could tell "I loaned X to Y" and such, be pretty easy to do if you had a device that could listen for an attention phrase, record 'til some amount of silence or something, then you could just transcribe and feed through an LLM with a system prompt to output such requests in database update-able form.
Sixty percent of freight containers or shippers have been canceled. Our whole industry has stopped ordering products from China due to the 145% tariffs.
and
These products will not be on the shelves because our industry and millions of small businesses have simply stopped ordering. We'll run out of inventory in the next 60 days.
Mack Trucks announces layoffs at Lehigh Valley plant, blames tariffs
“Heavy-duty truck orders continue to be negatively affected by market uncertainty about freight rates and demand, possible regulatory changes, and the impact of tariffs,” spokesperson Kimberly Pupillo said.
Both via this BlueSky thread.
NPR: A whistleblower's disclosure details how DOGE may have taken sensitive labor data from the NLRB. Always hard to tell what's a newsletter writer trying to pimp up mainstream recording, but this thread with screenshots from the testimony that say:
In the days after DOGE accessed NLRB's systems, we noticed a user with an IP address in Primorskiy Krai, Russia, started trying to log in. Those attempts were blocked, but they were especially alarming. Whoever was attempting to log in was using one of the newly created accounts that were used in the other DOGE related activities and it appeared they had the correct username and password due to the authentication flow only stopping them due to our no-out-of-country logins policy activating.
and that MFA got turned off.
Meanwhile, cyber professional @permadeath.com writes: "i am going to Lose My Fucking Mind if the seed crystal for an antifascist general strike ends up being david fucking brooks": NYT: What’s Happening Is Not Normal. America Needs an Uprising That Is Not Normal.
Thursday April 17th, 2025
So April 19th is "Bicycle Day", and Tara 🕷️ @[email protected]
someone at work just said something about "folks taking time off ahead of the holiday weekend" and i thought "wow, i wouldn't expect her to recognize 4/20 as a holiday"
... it's Easter. she means Easter.
Seeing a tour early bird special with "those who've already booked have had their balance adjusted", and now I'm wondering how many of those folks are stumbling!
On remapping Ctrl-c : "I can't imagine a reason that I would ever do this though".
That's because Julia is a *good person* who would never mess with, say, their coworkers who left a terminal unsecured. Or something. Hypothetically.
JA Westenberg @[email protected]
The most underrated cognitive bias: we intuitively understand that complex systems can't be controlled, yet demand that politicians promise to control them. Democracy then selects for the most convincing liars.
Gavin Logan @[email protected]
Looking at a tech conference that is "focused on sustainability" and will "heavily feature AI". Which is like being "focused on agriculture" and "heavily featuring locusts".
People talk about time travel to eliminate Hitler, and Nextdoor is right there...
Wednesday April 16th, 2025
Had to try it myself: indeed, Facebook search for "el salvador mega prison" brings up this warning about CSAM.
After that previous blocked one, we found the other chargers in downtown Hanford, plugged in, and have been wandering around... Come back and there's a police car parked in one of the charger slots, not plugged in. Kinda like it's municipal policy to just block the chargers...
OpenAI is building a social network. I don't remember what level of startup flail this is, I think it comes before cryptocurrencies and virtual reality, but honestly I can't remember where OpenAI is on those.
https://www.theverge.com/opena...enai-social-network-x-competitor
Tuesday April 15th, 2025
While I'm noting things for posterity, Charlene and I had coffee this morning with Dan Zack, talking about Fresno's transformation and urban planning generally. https://zackurban.com/
High speed rail is coming to the Central Valley...
Wandering around downtown Hanford. Proof that having preserved buildings from the 1800s doesn't alone make for a vibrant experience.
My earlier whine about how California needs an order of magnitude more charging stations didn't take into account municipal vehicles blocking chargers.
Monday April 14th, 2025
Ugh. Typo *and* no image. Trying again... Me too, Talbots. I'm petite inside. Right now pretty "bear" outside, though.
Me too, Talbot's. I'm petite inside.
Right now pretty "near" outside, though.
Hanging out in Fresno with Charlene's developmentally disabled brother, and the ways that he says things that we're never sure if it's perceptive or nonsense, and the way when we ask for clarifying information he says "yeah" and then comes out with a non sequitur reminds me of LLM output.
I'm in Clovis for a square dance calling gig. There are lawn signs advertising Easter at Cross City, and I have to imagine that this establishment is like Circuit City or Party City.
Or maybe Paradise City (where the grass is green and the girls are pretty)?
Sunday April 13th, 2025
Three weeks ago, California announced that it has more electric charging stations than gas station nozzles. So if one in three of those work, half of those are compatible with your car, and you can work out payment for 2/3 of the remainder, we've only got an order of magnitude left to go.
Saturday April 12th, 2025
Test fitting the banner for riding the quadricycle in the Petaluma Butter and Egg Days parade with the Safe Streets coalition.
In light of ad tech and AI making personalized fake relationships at scale easier: When a brand gets you to engage in an artificial parasocial relationship, is what they're doing any different than catfishing/pig-butchering?
Stored for dropping a link in the replies the next time some old high school classmate posts some anti-trans stuff on the social media. (I can't find it right now, which is why I'm saving this off, but it was some bigoted bullshit about about equating acceptance with bad parenting.)
The TGNB youth assigned male at birth with acceptance from at least one adult had 40% lower odds of attempting suicide in the past year compared with TGNB youth who were not accepted (aOR=0.60), and TGNB youth assigned female at birth with acceptance from at least one adult had 29% lower odds of attempting suicide compared with those who were not accepted (aOR=0.71). Acceptance from at least one peer was associated with 46% lower odds of attempting suicide in the past year for TGNB youth assigned male at birth (aOR=0.54) and 27% lower odds of attempting suicide in the past year for TGNB youth assigned female at birth (aOR=0.73).
doi:10.1089/trgh.2021.0079
Sam Altman: Three Observations
Still, imagine it as a real-but-relatively-junior virtual coworker. Now imagine 1,000 of them. Or 1 million of them. Now imagine such agents in every field of knowledge work."
The thing about inexperienced "junior" coworkers is that eventually they become senior coworkers. And the thing about work, is that the ratio of junior coworkers to senior coworkers is hopefully balanced to reduce the load on the senior coworkers.
Friday April 11th, 2025
Interesting to see Appellation Hotel flyers in downtown Petaluma. I've heard reports of merchants being afraid of the Nextdoor mob, I know it makes me more inclined to shop at places that I previously wasn't sure were really vested in the sort of vibrant down that I'm hoping for.
Of *course* building this sample code will require more time fucking with debugging node package bitrot than rewriting it myself.
The number of hoops people jump through to avoid typing a little bit of boiler plate...
We laugh about "2,000 years inventing written language and we're back to heiroglyphs", but I'm trying to imagine someone from the distant future, like next year, trying to understand what Unicode U+1F4F9 represents: 📹.
I suspect that only those of us from the 1900s have the context to understand.
It feels completely perverse to be using the Firefox translation feature for this, but the conclusions are so wonderfully blunt.
According to Business Insider, the new feature is not available in the free version of ChatGPT due to "copyright" issues, but it seems to be still available in the paid version. This feature has become very popular among people who don’t have a shred of imagination
Via.
I'm implementing a video chat feature this morning. I started out with a Medium post on WebRTC basic concepts and creating a simple video call app, but it turns out that that's mostly just a rephrasing and introducing of more NPM packages of Fireship WebRTC Video Chat on Firebase (to be fair, the former has a little more on building your own STUN server rather than using Google's), which itself is a distillation of the Fireship WebRTC Firebase demo code.
I apparently read very very quickly. Whenever I'm helping people with their computers, I have a long moment of "okay, we're looking for the button that says ... on it, so ..." and then a long detailed description of how they can find it as I watch their mouse cursor, when I absorbed the screen and located the action item sub-second. So I get why people are using LLM "summarization", but this comment spoke to me in the context of this morning while I'm digging through Medium/LinkedIn resume padding slop and thinking about how "AI" automates that on an industrial scale.
Ted McCormick @tedmccormick.bsky.social
Curtailing people’s ability to read widely and carefully, to locate, assess, and compare different sources for themselves, and to write in their own voice about what they find and what it means, is arguably more effective than censorship. It is also one of the most obvious effects of generative AI.
That this post makes a distinction between cars and "heavy machinery as some steam-powered orphan-eating contraption from the industrial revolution" says something about us.
We need to have the best tomato, raisin and vinegar technology in the world: The US Secretary of Education referred to AI as ‘A1,’ like the steak sauce.
In McMahon’s defense, it doesn’t seem like she actually thinks that artificial intelligence is abbreviated “A1.” During the panel, she said “AI” at first, but became increasingly less consistent.
“It wasn’t all that long ago that it was, ‘We’re going to have internet in our schools!’” she continued. “Now let’s see A1, and how can that be helpful.”
The matrix is perhaps glitching?
What do we call AI that works? Underpaid humans...
Edit: Pivot to AI on Nate.
Maybe the tide is turning on subscription software? PetaPixel: Adobe Deletes Bluesky Posts After Furious Backlash
Good dive into graphic design: Why do AI company logos look like buttholes?
Edit: Zack Whittaker @[email protected]
This has rewired my brain so now I immediately think AI stands for, "Asshole Inspired."
Summarized nicely by Kevin Beaumont @[email protected]
That report is wild btw, it’s basically ‘we’re going to set the planet on fire!’ along with ‘but generative AI will save the planet and cure cancer ‘cos Sam Altman is Jesus’