Show more

"We discuss the "Original Sin" of the Von Neumann architecture and how modern LLMs have made this problem even worse by completely erasing the distinction between code and data. From leaking private data to facilitating ransomware, we cover what a malicious prompt can force an agent to do and why standard defenses like "prompt classifiers" or "user confirmation pop-ups" are destined to fail."

youtube.com/watch?v=_3okhTwa7w4

Trump has been a deadbeat for decades, so no surprise that his regime is stiffing the UN on legally required dues payments, putting the entire organization into danger of folding.

apnews.com/article/united-nati

Todd C. Miller has been maintaining the #sudo codebase for over 30 years. This is exactly one of those cases where an entire critical infrastructure is held together by the work of a single volunteer who apparently can’t find anyone willing to sponsor him for some financial support. #opensource #linux #foss #GNU

Dear Fedi friends,

Good morning from #FOSDEM!

Yesterday was pure magic: I loved every single presentation in the Social Web devroom, with devs and Fediverse champions sharing their inspiring work.

I didn’t post anything because I was so absorbed in all the talks (plus @liaizon and @Sascha did a fantastic job chronicling the event).

I can’t stop thinking about one of @benpate’s closing slides:

“Every single thing you do to improve the Fediverse is a win for humanity”

YES! 🙌

@miramarmike in this case it does have to do with the particular person... but beyond that:
- people have always treated machines as a kind of oracle, even before LLMs
- people have always projected agency and intelligence on to computers, see early experiences with ELIZA onwards
- people feel that the computer doesn't judge so the interaction feels lower-risk...
- ... while at the same time people are poor judges of actual risks when it comes to computers
- LLMs have a "voice" that is co-operative, friendly and supportive by default
- LLMs give the impression that they "know" stuff -- they phrase responses as knowledge held by them, rather than (eg) "here is a response generated semi-randomly from your prompt words and a statistical analysis of a training set which may or may not contain relevant data"

A large measure of fault falls on the people designing these things. They needn't be engaging, friendly and reinforcing and promote more and more follow-up interactions, but that would not fit the commercial goals. Another measure of fault sits with people who do not understand how they work and who are insufficiently skeptical about [nearly everything], but who tell everyone else about LLMs (eg news media, influencers, ...).

I see it as a kind of hack of human psychology akin to how gambling works.

The Alex Pretti Memorial Ride in Minneapolis today was massive and so powerful to be a part of. I’ve seen estimates of 3-5 thousand riders, taking up to an hour for the ride to pass through an intersection.

Despite their masks, because they all BYO gear every one of them looks unique. It should be pretty doable to train a ML model to automatically cross-reference all the footage to build a profile on each of them. #ice

"[In a commons] ... there are engagements with the market, they're just not capital-driven, and extractive, and predatory.

...

There can be market activity, but as many Open Source communities know, introducing money and transactionalism to that community can poison relationships. Can start to erode the community. So there's a certain trickiness to dealing with money in a commons."

#DavidBollier, 2025

revolution.social/episodes/thi

#podcasts #NosSocial #RevolutionSocial #commons #OpenPlatforms

Java-GI 0.14.0 has been released!

Release notes: github.com/jwharm/java-gi/rele

Highlights:
- all bindings updated to #GNOME 49 version
- #JSpecify nullability annotations everywhere

For those who don't know, Java-GI is a #Java language binding for GNOME (and other GObject-Introspection libraries) based on the new FFM API in OpenJDK. It's modern, fast, and very easy to use. Visit java-gi.org for more information!

Save the Date! Vendor Applications are LIVE for the Third Annual Upstate Anarchist Book Fair on May 2nd and 3rd, 2026 in Binghamton, NY hosted at the PM Press warehouse. Learn more, join us, and apply soon at upstateanarchistbookfair.com/

$5 Weekend books are back! Today and tomorrow, the following titles are just $5, no coupon code needed. But we kindly ask that you please refrain from using any additional coupons/discounts along with this sale. We will be donating a portion of our sales this weekend to the legal defense of those arrested during the ongoing resistance to ICE. Shop at pmpress.org.

@luminos And I still want to see a network of cooperatives across Europe where Open Source developers can work together to create better solutions AND earn a living. Such an obvious thing to do, really, and so much more a practical approach than foundations.

Really, the only thing you need to know about Google's agentic #AI #Chrome "Auto Browsing" feature is that #Google says YOU are responsible for all actions the agentic AI takes on your behalf.

That's the whole ballgame. Right off the cliff. Don't touch it. Don't use it. Don't get anywhere near it.

The government’s arrests of journalists Don Lemon and Georgia Fort are naked attacks on freedom of the press.

Two federal courts flatly rejected prosecuting Lemon, but instead of accepting that humiliating defeat, the government has now doubled down.

Read our statement: freedom.press/issues/fpf-state

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