Show more

Cuadron and Gupta 2026: "Built on Broken Assumptions (BOBA): Why Coding Agents Cannot Be Left Alone" arxiv.org/abs/2611.02207 "On average [agents] complete 95% of requests taken one at a time, but when requests are chained together they complete 51%…The cause is the mistakes agents miss. Coding agents declare a request done once their own tests pass even when those tests miss a mistake. The next request then builds on broken code and fails about twice as often. The mistake is rarely found later."

Patel et al 2026: "Do Coding Agents Catch Their Own Bugs? (They Don't.)" arxiv.org/abs/2611.01181 "We evaluate eleven large language models and find that FRR ranges from only 6.13% to 24.53%, indicating that models largely fail to address problems introduced by the first PR change."

Well now I need to move off of Tuta too. Between their mealy-mouthed PR on genAI shit and the fact that their client is absolute dogshit, it's not a great experience.

Posteo doesn't do custom domains so that's out, unless I want to go through every damn service I use and change emails. Mailbox.org seems fine, I got a nice explainer from them on their genAI position and it's not as hardlined as I prefer though. Migadu maybe?

#tuta #posteo #mailboxorg #migadu

No matter how the elections turn out in the US today, at least I won't have to suffer through more inane political ads for a while now.

I'm pretty careful about curating media around me to filter out capitalist bullshit, but I've been avoiding watching anything on my free sub to Peacock particularly because of political ads. The non-political ones are bad enough, I can only watch short bursts of that service as it is. Luckily there's only a few things I actually want to see there.

Kind of annoying to have multiple things to be stressed about at once. Worrying about one thing at a time is more than enough work.

“The Endless Appetite for Fanfiction”

fansplaining.com/endless-appet

> In 2024, everyone wanted a piece of fic, from AI grifters to traditional publishers to ravenous audiences. Where did that leave the people who write it?

"Claude writes better code than I do"

Yeah, I'd keep that to myself if I were you.

“The Blackboard Way of Knowledge”

2ndbreakfast.audreywatters.com

> I find it useful to talk about history – in no small part because the tech industry seems to always wish we wouldn't. Perhaps because we'd ask things like "didn't you just promise to revolutionize education a few years ago with a totally different product?"

Still #amReading Welcome to #AuDHD by Megan Griffith.

In the section on how to cope with the executive function piece of social engagements, she mentions working backwards from when you need to be somewhere to create a schedule (ex: I need to be there by 8, it takes me 15 min to get there, 30 min to get ready, so leaving a buffer I need to start getting ready at 7).

I'm fascinated to hear what the alternatives are?
How does anyone ever get somewhere on time without doing this?

Mother Jones: Using AI to Plan Backcountry Trips Is Getting People Into Big Trouble (Nov. 1, 2026)

“I think they kind of turned their brains off.”

motherjones.com/environment/20

#SearchAndRescue #hiking #AI

I need a new job. Let me know if there’s an opening for a very thoughtful, playful person who despises AI, cares about web standards, actually loves CSS and UX, has a long history in animation, and can write web or game code that ships with remarkably-few bugs (because I’m persistent and attentive)

Show thread

As an admin on Fedi, I used to welcome new users, fight Nazis, and suspend trolls.

Now all I do is fight against the endless army of soulless scam machines, forever marching against our server walls in an attempt to destroy the community we've built here. It's exhausting and disheartening. At least the Nazis and Trolls were real people who would get pissed and fuck off eventually, but the LLMs will run as long as someone pays the electric bill for them.

Show thread

The term "cargo cult" comes from a racist story popularized by a sexual predator; I'll say that the AI industry is rife with mimetic isomorphism instead.

Not only are LLMs themselves automatic devices for imitating form without function or purpose, but you keep running into shit like this, where the AI industry forms "institutes" with "scientists" and "researchers," all imitating the form of academia but not its function of inquiry or curiosity.

wandering.shop/@theodoraward/1

I am extremely hopeful about the outcome of America's elections tomorrow. I'm also extremely skeptical about the polling numbers I'm seeing. Yet a friend with deep knowledge says that -- this time -- the polls are right. Doesn't matter. Vote as if your life depended on it. Because it might.

Forgejo will be removing support for the OpenID Authentication 2.0 standard for authenticating users in 2027. Note that OpenID Connect (OIDC), and OAuth are unrelated technologies that remain supported. For information on the specific dates that support will be dropped, please refer to: forgejo.org/2026-11-removing-o

Reminder: that chat bot service you’re using may die when the AI bubble pops. If they do, they will likely sell all of the user data.

Read the terms, read the privacy policy, and still be very careful what information you give out. You will almost never get a say in who gets to buy that data when the company goes under.

Get AI agents the fuck off Fedi!

They're an existential threat to small instances, and they're a primary contributor to moderator burnout.

#FuckAI #IfYouSeeItReportIt #AI #LLM #NoBots

Show more
Librem Social

Librem Social is an opt-in public network. Messages are shared under Creative Commons BY-SA 4.0 license terms. Policy.

Stay safe. Please abide by our code of conduct.

(Source code)

image/svg+xml Librem Chat image/svg+xml