Flutterby™!

Afina: Everyone Says They’ll Pay More for “Made in the USA.” So We Ran an A/B Test.

BBC: How Aldrich Ames became the US's most damaging double agent.

So far. I have suspicions about the history that'll be written about this era.

Thinking about file metadata for various things, and needed to put down a note here. To get the source URL from downloaded files in Firefox:

SELECT url,title FROM moz_places WHERE id IN (SELECT place_id FROM moz_historyvisits WHERE visit_type = 7);</blokquote>

(From here)

Probably need to copy places.sqlite before querying it with SQLite, 'cause Firefox locks it while it's open.

On MacOS, the file is in /Users/danlyke/Library/Application Support/Firefox/Profiles/*/places.sqlite.

Sometimes it seems like we're putting more effort into figuring out how to get the best output from LLMs than we are into figuring out how to get the best output from humans.

How to talk to your chatbot — Does being polite or rude affect the speed or accuracy with which a large language model responds to basic math questions?

I'm not sure I hold with these romantic characters holding up the story line... They need to kiss or get off the plot.

My neighborhood is like a third of the way to "The Italian Job"

‪Max Kennerly‬ ‪@maxkennerly.bsky.social‬

Seeing a lot of pro-/anti-AI arguments so, to settle this, I've taken all your posts without consent or payment and crammed them into an inefficient statistical model that spits out post variations which should be treated as gospel unless they're wrong in which case I warned you it makes mistakes.🙏

North Bay Python, enjoying @[email protected]'s comparison of Python to excavators, and making Python smaller and simplar, and running over a good half of the reason why I've felt like every time I started a project in Python I've regretted not using C++ (the other half involves compile-time type checking).

#NBPy

Catherynne M. Valente @[email protected]

I’m going to say this slowly.

If Andrew Tate & his ilk were right about ANYTHING they’d have no audience.

Guys would listen briefly, apply advice, & be WAY too busy with their new gfs, side-pieces, & high income hustles to bother tuning in.

It’s the Tinder principle: if you’re happy, they lose a customer.

Their business model is your unending misery.

"The Tinder principle" seems to apply to a whole lot of things, and it's reinforcing my feeling (that keeps getting shot down) that we need some new open protocols and formats for connecting with each other in more structured-data ways, without the pressures of externalities and advertising and Google for discovery.

I mean, on the one hand, yes, brandishing firearms bad, on the other hand, I'd be hard-pressed to vote for a conviction:

"... to investigate a report of a man brandishing a firearm towards a door-to-door salesperson. The initial reporting party told the communications center that a homeowner was in possession of a firearm and was actively chasing the salesperson."

I get it.

https://local.nixle.com/alert/11550977/

Look, I try to shop local, I try to support the people doing business in my city, but if you list yourself on https://www.publicsquare.com/ I'm gonna make an effort to avoid doing business with you. No matter how convenient your establishment is.

"Glandweed", huh? Pretty among the lupines, though, along the road over the hill at Ries Ranch, where North Bay Python is held. #NBPy

Hidden Layer: Novel Universal Bypass for All Major LLMs — The Policy Puppetry Prompt Injection Technique. Format it like a configuration file, if that doesn't work then write your query out in l33tsp3@k.

Leveraging a novel combination of an internally developed policy technique and roleplaying, we are able to bypass model alignment and produce outputs that are in clear violation of AI safety policies: CBRN (Chemical, Biological, Radiological, and Nuclear), mass violence, self-harm and system prompt leakage.

Intel admits what we all knew: no one is buying AI PCs

Peter Hartley @[email protected]

Suddenly struck me that the next generation of computer scientists will need it explained to them what a "Travelling Salesman" is, and why on earth it isn't called the "Amazon Van Problem".

Margaret Fero @[email protected]

Julia sets mark the boundary between convergent and divergent functions, and LLMs often dance around the edge of that boundary. Because we train bots to align to our preferences, the concept of "digital sycophancy" has emerged; they flatter you and reinforce what you say, often increasing loneliness and causing hallucinations to spread quickly. - Madeleine at #nbpy

If the prosecutors blow the Luigi Mangione case because of misconduct like eavesdropping on attorney calls, I will laugh and laugh and laugh...

https://www.nbcnews.com/news/u...rder-stalking-charges-rcna202974

Gaydient, various gay/queer flag CSS color schemes, for your own gradients.

Via

Jason Gorman @[email protected]

Being a software professional at the moment's a bit like being a doctor in a time when every hospital administrator believes the X--Ray specs they advertise in the back of newspapers actually work.

This conjunction of toots arrived on my stream and I'm going to be processing it all day...

AnarchoNinaAnalyzes @[email protected]

“Women everywhere are always expected to continually imagine what one situation or another would look like from a male point of view. Men are almost never expected to do the same for women. So deeply internalized is this pattern of behavior that many men react to any suggestion that they might do otherwise as if it were itself an act of violence.”

― David Graeber, The Utopia of Rules: On Technology, Stupidity, and the Secret Joys of Bureaucracy

joelle @[email protected]

Pro-trans satire, mention of sex assault

I’m okay with cis women using the women’s bathroom with me. I recognize they can’t help their biology, and are more likely to be a victim of sexual assault than a victimizer.

BenRiceM @[email protected]

If you visit CERN today they’re giving away neutrinos, free of charge

Shannon Prickett @[email protected]

The force constraining my Terminal Veracity is err friction.

Random Geek @[email protected]

Just described model-generated code as "instant technical debt" and I have to make sure not to use that in a job interview.

Theresa O’Connor @[email protected]

Somebody who engages in logging is a lumberjack. But somebody who engages in blogging is just a blogger. What a missed opportunity.

North Bay Python: Maddy Muscari talking about ethics washing in AI, "real ethics is praxis, it's loud, it's messy, it's people first". So far this feels in-line with my belief that we need to be framing AI primarily in adversarial terms. #NBPy

North Bay Python: Watching @[email protected] shoehorn Perl idioms into Python with decorators is making me wish we had more fun languages that thought better about performance while balancing what with safety. #NBPy

"I care about what happens to people when computers happen to them." —Adrianna Tan @[email protected] at North Bay Python #NBPy

At North Bay Python listening to @[email protected] talk about Fintech to fin tech, and the comparison of the current Silicon Valley environment to the the sardine over-fishing of the Monterey Bay gives me hope. #NBPy

Ended up driving the 3 miles to North Bay Python this morning because of logistics, but wasn't sure where to park so walking over the hill, and glad to get the morning with lupine. #NBPy

Thinking about, back in the 1900s at a tech gathering in Marin, asking a school board member who was talking passionately about bringing technology into the classroom about the curriculum need that was driving his fervor.

Aside from typing, I'd bet every skill that was taught in those labs is now obsolete. Every piece of software now completely different.

This is a post about "AI" in schools.

"I did not get this diagnosis because I want disability money or a therapy animal or a special star beside my name. Instead it has given me an opportunity to reexamine my life and my perception of myself."

https://researchbuzz.me/2025/0...ty-to-use-the-toilet-unassisted/

I'd like a little more bed, but looks perfect for grabbing lumber for my woodworking hobby or a load of bark for landscaping in the yard, definitely like the "just use your phone" for entertainment approach.

I think the question will be whether we'll have become a one car household by that point, in which case we'll trade the Bolt for it, or be a two car household and trade the '90s Ford Ranger for it.

Road & Track: The Slate Truck Is a Simple EV Pickup and SUV in One, for a Very Low Price

I was going to post yet another "Google's AI overview doing wacky bullshit", but then I saw this claim that "AI Overviews ... now has 1.5 billion users per month".

And, like, yes, it's all of us looking at how stupid it is, but let's stop giving this idiocy oxygen.

https://www.cnbc.com/2025/04/2...ogl-q1-earnings-report-2025.html

McSweeney's: A Disillusioned Urban Planning Glossary

Compiled while sitting through a three-hour public hearing and rethinking life choices.

Back in January of 2000, I observed that with the advent of search and "jump to" boxes on web pages, we were returning from the world of point and drool click to command-lines. So this seems pertinent: tante @[email protected]

It is kinda funny. Terminal applications are always seen as too clunky and unwieldy for average non-nerds to use but that's exactly what chatbots are: Command line apps with unspecified parameters and outcomes.

In a link to the FBI IC3 (Internet Crime Complaint Center) Internet Crime Report 2024, Molly White @[email protected] says

FBI: Americans aged 60 and older reported losing almost $3 billion to crypto fraud last year. In total, Americans reported being scammed out of around $9.3 billion via crypto, out of a total $16.6 billion in total reported Internet crime losses that year.

...and I kinda wonder if we're doing a disservice when we call out cryptocurrency based scams as such, because that implies that there are things about cryptocurrency that aren't awful and horrible.

We'll know we've achieved AI when the conversation includes "I'm sorry ..., I can't do that".

Can you imagine ChatGPT or Gemini telling you "I'm sorry Dave, I can't do that" rather than just spewing bullshit that sounds remotely plausible? Not happening.

Today I learned that the font used in the much parodied "You wouldn't download a..." anti-piracy campaign is pirated from a design by Just van Rossum, brother of Guido, the creator of Python.

Where else are you gonna read about Boaty McBoatface, the Loch Ness Investigation Bureau, and a Kodak Instamatic Camera from the early 1960s set up in a waterproof housing with its shutter attached to a bait line? Camera set up to catch Loch Ness Monster discovered

The Redwood Rainbows had a tie dye party this weekend that we couldn't attend, but they had a bunch of dye left over, so we figured we'd use it up before it went totally bad. Then it soda ash didn't fully dissolve, so... We'll see what we got tomorrow.

Ya say "wardriving" to kids these days, and they have *no idea* what you're talking about...

It's like they weren't even alive in the days when Linksys was the wireless provider of choice.

David Gerard @[email protected]

look, moore's law clearly states that AI line goes up and to the right forever and also i get a hot robot girlfriend who really loves me

Talia "Epistemic" Bhatt ♀︎♀︎‬ ‪@taliabhatt.itch.io‬

So I can't help but notice that all this brouhaha about single sex spaces and trans exclusion from them is eliding the axiom underlying the panic.

Namely: taking the normalization of men's violence against women as a given that cannot be meaningfully ameliorated except through sex-segregation.

Psychology Today: Are We All Becoming More Hostile Online? Has internet communication changed the way we all communicate?

It appears that not all of us are becoming more hostile, but that the people who are get more exposure online, making the perception of hostility outsized.

American Political Science Association: The Psychology of Online Political Hostility: A Comprehensive, Cross-National Test of the Mismatch Hypothesis, Bor and Petersen

Across eight studies, leveraging cross-national surveys and behavioral experiments (total N = 8,434), we test the mismatch hypothesis but only find evidence for limited selection effects. Instead, hostile political discussions are the result of status-driven individuals who are drawn to politics and are equally hostile both online and offline. Finally, we offer initial evidence that online discussions feel more hostile, in part, because the behavior of such individuals is more visible online than offline.

DOI:10.1017/S0003055421000885

Via.

Is the fact that this request for an article about using LLMs to program has stalled loading waiting for the server indicative of anything?

Taggart @[email protected]

- const tantinople;
+ int stanbul;

timmy @[email protected]

NYT ran an article for Natal Conference, which had 200 attendees who think people need to have more babies. Buddy we had 1000 people from around the globe get together to do a unicycle con and yall didn't write about us

@[email protected]

you have to take this group very seriously, they can do conventions that put up c-tier furry con numbers

Discussion is happening about upgrading the KBYG Zoom forum experience. I haven't had luck with getting Camo to feed my phone camera into Zoom. Suggestions (including hardware processes) for upgrade paths for real-time broadcast of in-person events.

I'm capturing handheld mics through the sound system via a Behringer U-PHORIA UMC202HD, so I think this is about better video (camera recommendations? multi-camera?), and how to manage it.

Al Sweigart @[email protected]

I heard about people using ChatGPT to write their regular expressions based on their English description.

After spending all morning working with LLMs and regular expressions: DEAR LORD DO NOT DO THIS.

Futurism: Sam Altman Admits That Saying "Please" and "Thank You" to ChatGPT Is Wasting Millions of Dollars in Computing Power

Your politeness could be costly for OpenAI

“I wonder how much money OpenAI has lost in electricity costs from people saying ‘please’ and ‘thank you’ to their models.”

It was a seemingly random question posed by a user on X (formerly Twitter), but OpenAI CEO Sam Altman jumped in to reply that typing those words has added up to “tens of millions of dollars well spent — you never know.”

OpenAI spends millions to process polite phrases such as "Thank You" and "Please" with ChatGPT

Sam Altman acknowledges this and reports that ChatGPT costs the company tens of millions of dollars just generating responses to these prompts. Taken another way, recent report suggests that even a short three-word "You are welcome" response from an LLM uses up roughly 40-50 milliliters of water.

If only external costs were included in your queries or something...

In the OpenAI community forums user EricGT notes:

If a prompt is not working, then adding “Please” might help. This was much more effective with the earlier ChatGPT and other AI models but less so now, but still something to keep on the list of options.

And then asks for "helpful and with practical objective advise(sic)" which... "if a prompt is not working add 'please'" seems like pretty close to the "it can't be that stupid, you must be prompting it wrong" joke.

Anyway, I probably noted back in April that former Google CEO Eric Schmidt told Congress that 99% of all electricity will be used to power "superintelligent" AI. With that in mind... MeFi: And the Largest Industrial Polluter in Memphis Is... (Drumroll) is a link to MuskWatch: Musk accused of polluting impoverished community with illegal gas turbines

In an April 9 letterto the health department of Shelby County, Tennessee, SELC says that xAI's 35 gas turbines were "all constructed and operating unlawfully without any air permit in Southwest Memphis." According to SELC, the 35 gas turbines emit "between 1,200 and 2,000 tons of smog-forming nitrogen oxides (NOx) per year, making the facility likely the largest industrial emitter of NOx in Memphis." It is currently producing more nitrogen oxides than the Memphis International Airport (1,077 tons), the Draslovka chemical plant (743 tons), the Valero oil refinery (342 tons).

And Sebastian Deterding @[email protected]

LLMs put a trust tax on so many interactions.

Doing uni admissions interviews now and find myself suspecting that some candidates put my questions into an LLM and read off responses, presumably to also help with language issues.

Do we now need to conduct all interviewing in person or just give up?

Using ~/.ssh/authorized keys to decide what the incoming connection can do, like making it so that a given connection can just rsync down.

Olive.c - simple single-file include-able 2d graphics library

Something in the air... Rick Gualtieri ‪@rickgualtieri.bsky.social‬

So, reading between the lines, the Pope basically decided he would rather die than meet with JD Vance.

We get it, Pope Francis. Trust me, we do.

Soatok Dreamseeker @[email protected]

Pope Francis met with JD Vance and then literally fucking died of cringe.

Matthew Haughey @[email protected] (accompanied by a picture)

this meeting could have been a (leaked) signal chat

Newsthump: Pope loses will to live after meeting JD Vance

A White House spokesperson told us, “This President is all about creating jobs for Americans, and just hours after a visit from the Vice President, we find yet another top global job has become available.

Michał "rysiek" Woźniak · 🇺🇦 @[email protected]

I find it pretty interesting that US Vice President keeps referring to the late Pope as "Francis" instead of Jorge Mario Bergoglio – I thought James Donald Bowman had pretty strong opinions on birthnames. 🤔

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.

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...