V roce 2025 se v AI kodovani ujal "deep modules" (simple interface, complex functionality) jako zasadni pattern. V podstate cele Unreleased je takhle napsane.
Rok 2026 ukaze, jak je to udrzitelne. Kod v modulech je v podstate "write once, edit never". Kdyz je neco potreba, cele se to prepisuje. A API kazdeho modulu hlidaji agenti, aby se pouzivalo a neduplikovala se funkcionalita.
Rikate si, ze takovy AI agentni system poroste exponencialne?
Ano.
Bude to zabava.
In all the time I've had my Zoom H5 Studio, I've really never given it a good SPL tolerance test. So, today, I went out to the nearby train tracks, and recorded two trains, both with the includec X/Y microphones, and a pair of Micboosters Clippy Pro.
The Clippy Pros lived up to their name, and the recording I got from them was basically unusable distorted bad. Next time, I'll try turning the pads on the inputs and see if it's the mics or the recorder that didn't like things.
Anyway, here's one of the two trains I managed to capture. This one is fast and short. I didn't use any processing, but I did ride the gain a little bit to make it a little easier to deal with the wide dynamic range. If anyone wants the raw, let me know.
For best affect, play it loud through something with lots of range.
Every day I see some horrifying thing (today it's ICE killing someone in Minneapolis) and then I have to look at my work computer like "okay let's review this cute little design here" like nothing is happening.
Sometimes the routine comforts me. Other times, it feels absurd and painful.
Totally normal HN poster about LLM running on a raspberry pi
> We're approaching the point where we could have sophisticated sound controlled sex toys.
Wow, relying on Signal might actually be a Very Bad Idea™
In below longread there's a lot to unwind, but the essence is this: it is a state asset for American imperialism built on the infrastructure of Big Tech.
So... uh... @delta it is then?
counterpunch.org/2025/03/07/th…
We are constantly told that the messaging platform Signal is totally secure and benevolent. While Signal may be preferable to the dominant alternativesThe Mapping Project (CounterPunch.org)
@feld I think we have diametrically opposite preferences for how to manage things. I find that having stuff install outside of /usr/local and without using pkg or ports is generally a bad idea. If I need stuff from pip/npm/whatnot, it should be installed locally to a given user, not globally like suggested here. My biggest concern is the lack of maintainability.
But I'll see if I can give this a go, somehow. There's nothing in this bundle that, apart from some custom-built packages, should need anything special, so converting to a base-package approach should be mostly a matter of "getting it done".. :)
@mischievoustomato @ltning for certain automation/monitoring tools you either have to break the unix mantra of "everything is shared, only one copy of stuff installed" or build static binaries
that's another reason why Go got so popular -- it avoids this mess by producing mostly static binaries
@mischievoustomato @feld I don't quite understand this concern. If stuff is broken in the OS package repo then you're screwed anyway and I certainly wouldn't trust a config mgmt tool to "fix" anything. But I'm sure there are pains I have not experienced, so what do I know ;)
And regardless of all this: chatrelay depending on custom patches to upstream software is bad enough; that it seems hard-to-impossible to manage it using standard tools (even on the recommended platform) seems to me to be unfortunate. Each individual part looks fairly straight-forward from a cursory glance..
All that said: @feld - thank you for making the effort. You've made it much easier for me to understand how it all works, and potentially getting my own relay off the ground.
> chatrelay depending on custom patches to upstream software
It's actually zero. The only patch they have is a dovecot change that removes an unnecessary sleep/debounce that slows down message notification/delivery by 500ms
github.com/dovecot/core/pull/2…
everything else is just the specialized configuration and some custom python (later: rust) services that filter emails, lua scripts for dovecot and opendkim.
> If stuff is broken in the OS package repo then you're screwed anyway
I think FreeBSD believes the opposite. It's why we have a /rescue. 😀
The U.S. this morning seized oil tankers in international waters... just south of ICELAND.
Nothing to do with drugs.
Nothing to do with security.
Just straight-up theft.
If you think the U.S. does not plan to take Greenland, Canada, and other countries (by military force if it deems it "necessary"), you are now living in a fantasy world.
STOP 👏🏻 BUYING 👏🏻 AMERICAN 👏🏻 PRODUCTS
(Yes, this is a real tweet from the U.S. State Department. See for yourself: x.com/StateDept/status/2008221…)
I get that same uncomfortable look when someone tries to hand me a baby every time someone gives me a shoutout for some open source thing
"eehhhhh this is too big of a responsibility, i really shouldn't. no, you don't understand, you don't want me in this position it may not end well"
@mischievoustomato I can't call myself a programmer, never really been paid to "build something" as my primary job description.
I'm a sysadmin/network guy who has been up to his eyeballs in Linux/BSD for a long time and spent a lot of time reading code and fixing things.
I'm not fluent in PHP, but I've fixed customers' PHP apps before.
I've reworked a patches to C/C++ projects to make them compile again on newer OS or library or whatever
Give me a targeted error or problem and I can probably figure it out.
Ask me to build/design something from scratch and even though I have a lot of ideas in my head about the right way to do things because I try to pay attention to best practices, security, performance and have a pretty wide range of knowledge... I'm more likely than not to produce an amateurish pile of garbage
If someone else provided me with a complete technical design to follow of how it should work, there's a much better chance I can be successful or at least not waste so much time
Hi there! I see over in Disable primary-paste by default (!119) · Merge requests · GNOME / gsettings-desktop-schemas · GitLab that Gnome is considering removing middle-click paste.GNOME Discourse
OK so I've been talking about The End of UX for a while now.
I want to shoutout the @firefoxwebdevs account for *not* ending UX. They are engaging with some pretty heated feedback regarding LLM-related UI and having a genuine dialogue. I may be pretty much on the side of the heated feedback, but I also want to give credit where credit is due.
The End of UX would be to say "we'll design this however we want and you'll just love it” rather than engage. FF isn't doing that, which is commendable!
Put down toast. Barely toasted. Ok, let's toast one more time.
BURNED.
how does this happen, science ?
Bearded dog hazards:
Making breakfast, she drinks a bowl of water and runs over to me -- dripping water everywhere
I set down my egg and grab a towel
SPLAT
egg rolled off the counter
Here’s our summary of updates, events and activities in the LibreOffice project in the last month of last year – click the links to learn more… At the start of December, we announced a new Code of Ethics and Fiduciary Duties for The Document Foundati…Mike Saunders (The Document Foundation)
Tamas G reshared this.
It is absolutely *wild* to me that media organizations still do not put RSS/Atom 
feed info front-and-center on their websites.
One needs to dive into HTML code or use external tools to discover their feeds.
It is wild because this is one of the easiest, least-effort ways to reach their audience. Encouraging RSS/Atom use is a phenomenal way of becoming less reliant on gatekeepers like huge social media platforms.
Come on! 
If you're wondering what consequences X has faced: none. At all. A few months ago when Grok called itself MechaHitler, the service was shut entirely for days, the same day. When this women issue happened, Elon laughed.
Grok is still outputting non-consensual deepfake pornography and sexual abuse material at a rate of 1 post per second. Example search:
from:@grok filter:media
Direct link:
x.com/search?q=from%3A%40grok%…
I hope Cisco enjoy directly funding sexual abuse material.
It hands them blueprints for exploitation, legitimacy and controlchrisyoong.com
Putting people in bikinis is just the tip of the iceberg. On Telegram, users are finding ways to make Grok do far worse.
sightlessscribbles.com/the-hom…
It's so easy to forget how beautiful words can be. Thank you for the reminder, @WeirdWriter.
Everyone else: I would highly recommend his writing! See sightlessscribbles.com (there is an RSS feed).
I have a product idea that can save people money, give them their attention back, bring people closer together, divest in evil corporations, and help maintain healthy cognition/media literacy.
I call it “reading books and going to libraries”
Just figured out (out of necessity) that I have a better way of OCRing my vm's screen than using Seeing AI. If I specify a monitor when starting qemu, then I can telnet into the monitor, and there's a screendump command that will dump the screen to a file, and then I can use tesseract to OCR the image.
Unfortunately, none of that is helping me right now. I get "Guest has not initialized the display (yet)." So I'm just as mystified as before. Reminds me of the children's song I used to listen to where a climber went over the mountain to see what he could see, but the other side of the mountain was all that he could see.
Now I have this function in my .bashrc (I have the monitor on port 4444):
ocrvm()
{
rm -f a.png a.txt
echo screendump a.png |nc -N localhost 4444 >/dev/null
tesseract a.png a >/dev/null 2>&1
more a.txt
}
It isn't very polished--it creates files called a.png and a.txt in the current directory, but, meh, it does what I need.
Edit: I found my original issue--something was wrong with the virtual drive that I have set for the UEFI image. And I love having this little OCR scriptlet now--it makes it really easy to monitor the status of a vm while it's booting up and I don't have a screen reader available yet.
Khronos reshared this.
André Polykanine reshared this.
OneDrive is Microsoft's cloud-based storage service, and it's aggressively pushed on users of Windows. They want your files on their servers, and are willing to use dark patterns to get…Rob Beschizza (Happy Mutants, LLC.)
reshared this
#XSF Announcement
The XSF is considering to participate the #Google Summer of Code 2026!
If you are interested as a contributor or mentoring #XMPP project start reading here:
wiki.xmpp.org/web/Google_Summe…
#GSoC #chat #messaging #jabber #standards #opensource #interoperability #rtc #specifications
Hey @FreakyFwoof, I've just started this one. Interested to know if you think it's worth carrying on!
Harry Potter and the Hero's Path, By TheJackOfDiamonds
AU Ritually abused by the Dursley's, young Harry Potter learns to count on himself. After discovering magic at a young age, he practices to become stronger to protect those weaker than him.
March 2006-November 2025.
323493
words in 52 chapters.
fanfiction.net/s/2869936/1/Har…
johann reshared this.
Martin Wenisch
in reply to Martin Wenisch • • •Ted si asi tri dny hraju se ciste agentnim projektem. Rozrostl se mimo hromady balastu na 10 stabilnich modulu. Pri kazdem implementovanem "ticketu" kontroluji AI agenti QA tech modulu.
Pouzivaji se v novem kodu spravne. Neduplikuje se jejich funkcionalita. Je potreba nejaky modul upravit => je jeho API upravene optimalne.
Samotna agentni kontrola modulu (~22k radku Rust kodu) sezere 100% limitu nejvyssiho tarifu Claude Code.
Uz ted takhle naivni pristup nema kam rust.