mastodon - Link to source

Yeah that's fair. I have just read it's a 1.2 trillion -parameter model, that's quite insane. I wonder though if Google will update that as often as their own models, it might be that they still release newer models for themselves but then strategically hold the Apple one back a little bit, on the basis of "letting that model mature first." This will be interesting to see play out for sure.

dragonscave.space/@miki/115895…

miki

2026-01-14 21:29:43

mastodon - Link to source

Nah. Giving LLMs their identity is done in post-training, and Google can just remove that data for their posttraining mix for Apple. They're definitely going to have one. Since pretraining = ~60-90% of compute, redoing that part won't be that expensive. There are rumors that Google will only ship the base models anyway, and that all the fine tuning will be done internally by Apple (perhaps on Google infrastructure). Those base models have no idea what they are yet.

mastodon - Link to source

The most painful lesson I learned, and have yet to fully recover from, after losing the community I spent years cultivating when Twitter was sold, it’s this: the “revolution” and change so many of you say you want will never be built in public, on platforms owned by people who not only do not share your values but who see them as threats.

At any moment, the communities, businesses, networks, and even friendships you’ve built there can be taken from you because they were never truly yours.

Build Where Your Values Can’t Be Deplatformed

mastodon - Link to source
I retrieved this from David Goldfield's Email list. This client shows an enormous amount of promise and I am already liking it after five minutes of use. Quoted text:
My new Mastodon/Bluesky client, FastSM, which is based off the old work from me and @TheQuinbox, is now available in really, really early alpha! You can
download the latest build, which gets updated with each new commit, at
github.com/masonasons/FastSM/r…
@Quin

hometown - Link to source
uspol, ai software dev

Sensitive content

in reply to allison

mastodon - Link to source

miki

uspol, ai software dev

Sensitive content

mastodon - Link to source

Want odds on who/what blocks our driveway first?

A snowplow pushing the snow from the bus stop more than 1 block away, or a motorist who gives zero f*cks about blocking our driveway w/ their vehicle whilst they do whatever they need/want to do in our immediate 4-block radius?

Montréal braces for heavy snow as Environment Canada issues warning... montrealgazette.com/news/weath…

#winter #winterweather #CDNNDG #CDNNDGBS #polMTL #MTLpoli #meteoQC


Montreal braces for heavy snow as Environment Canada issues warning

Environment Canada has issued a warning for the area, saying rain is expected to turn to snow.
Montreal Gazette
#winter #polmtl #mtlpoli #winterweather #cdnndg #cdnndgbs #meteoqc

pleroma - Link to source

so this guy claims that he was was a sex slave for Epstein & co as a kid and Trump was one of his abusers (also Clarence Thomas and some others) and that at one point he jammed a tent stake up Trump's ass and got beat for it

he says Trump's medical records will corroborate this and this is also why Trump said he "felt raped" when his medical records were stolen a while back?

WILD SHIT MANNNNN :stonklol:

mastodon - Link to source
Oi, back to Tweesecake / TWBlue for now. Gosh. Looks like FastSM doesn't use the straming API yet so my Mastodon server keeps e-mailing me that people have mentioned or boosted things. Oof my poor inbox. (as a side note it's not the biggest deal as these can be turned off in prefs, but still.)
Good news, it can! You just have to enable that in "advance" settings to be used for home and notifications timelines.
This entry was edited (2 days ago)
in reply to Bri🥰

mastodon - Link to source

Tech Singer

@Bri @jonathan859 It's great work as always. Three small things, if I may raise them. First, I can't seem to move through threads on bluesky. Is this expected? Secondly, is it possible to have it say how long ago a post was made, rather than state it's time? I removed that from the template as it stands, would it be possible to say "1 hour ago" or something like that? Finally, this is more stupid user than even just a feature, but is there a keystroke to move from account to account in the invisible interface?
@Jonathan @Bri🥰
in reply to Bri🥰

mastodon - Link to source

Tech Singer

@Bri @jonathan859 One more quick thing. Are you supposed to be able to select any of the users in a post? when I hit, for example, the "follow user" key, I get the first user, the one who posted, but not the second user who the first user was responding to or boosting. Up and down arrowing doesn't get me to that second user as in TC. BTW, may I congratulate you for using the TC command structure? It makes life so much simpler.
@Jonathan @Bri🥰
in reply to Bri🥰

mastodon - Link to source

Tech Singer

First, I really like the open URL bringing up a dialog with the thing already listed. This is seriously helpful for those of us using speech/braille, a benefit for security, and generally a major benefit. Thanks. I don't think it has anything to do with the issue I'm seeing, but I may well be wrong. Sometimes, the field is blank and there is no url. When it hit OK, nothing seems to happen. I'm not sure where the problem happens, but I think it has to do with the post and not the timeline because I've seen it in many timelines.
This entry was edited (2 days ago)
in reply to Bri🥰

mastodon - Link to source

Tech Singer

@Bri Sorry to disturb but I recently lost connection for a while and, rather than silently stopping, got the below. Nothing has changed with the BS account, but I'm getting these errors
Error in notifications: InvokeTimeoutError
Error in mentions: InvokeTimeoutError
Error in search: InvokeTimeoutError

I looked in the errors.log and just got this, which seems to have nothing to do with the above errors. Note that I don't see a time stamp on these at all, I've copied the whole file which is 734 characters. Also, I wasn't doing anything with the keyboard, it just started speaking the above errors, I was working on another machine at the time and hadn't touched the keyboard for at least an hour. Would it be possible for the thing to say "can't connect, will do so in five minutes", or just not to say anything and keep trying every little while?
Traceback (most recent call last):
File "keyboard_handler\wx_handler.py", line 75, in <lambda>
File "keyboard_handler\wx_handler.py", line 105, in process_key
File "keyboard_handler\main.py", line 57, in handle_key
File "GUI\invisible.py", line 102, in speak_item
IndexError: list index out of range
Traceback (most recent call last):
File "keyboard_handler\wx_handler.py", line 75, in <lambda>
File "keyboard_handler\wx_handler.py", line 105, in process_key
File "keyboard_handler\main.py", line 57, in handle_key
File "GUI\invisible.py", line 178, in next_from_user
File "GUI\main.py", line 514, in OnNextFromUser
File "GUI\misc.py", line 606, in next_from_user
IndexError: list index out of range

@Bri🥰
in reply to Bri🥰

mastodon - Link to source

Tech Singer

@Bri I recently upgraded to the latest build and it's not finding anything for searches on BS, I'm afraid. My network connection is fine, but I'm getting "Error in search: InvokeTimeoutError". All I'm getting as errors in errors.log is Traceback (most recent call last):
File "keyboard_handler\wx_handler.py", line 75, in <lambda>
File "keyboard_handler\wx_handler.py", line 105, in process_key
File "keyboard_handler\main.py", line 57, in handle_key
File "GUI\invisible.py", line 164, in top_item
File "GUI\invisible.py", line 81, in focus_tl_item
IndexError: list index out of range
Traceback (most recent call last):
File "keyboard_handler\wx_handler.py", line 75, in <lambda>
File "keyboard_handler\wx_handler.py", line 105, in process_key
File "keyboard_handler\main.py", line 57, in handle_key
File "GUI\main.py", line 409, in ToggleWindow
wx._core.wxAssertionError: C++ assertion ""N == (-1) || IsValid(N)"" failed at ..\..\src\msw\listbox.cpp(356) in wxListBox::DoSetSelection(): invalid index in wxListBox::SetSelection
Traceback (most recent call last):
File "keyboard_handler\wx_handler.py", line 75, in <lambda>
File "keyboard_handler\wx_handler.py", line 105, in process_key
File "keyboard_handler\main.py", line 57, in handle_key
File "GUI\invisible.py", line 164, in top_item
File "GUI\invisible.py", line 81, in focus_tl_item
IndexError: list index out of range
Traceback (most recent call last):
File "keyboard_handler\wx_handler.py", line 75, in <lambda>
File "keyboard_handler\wx_handler.py", line 105, in process_key
File "keyboard_handler\main.py", line 57, in handle_key
File "GUI\invisible.py", line 164, in top_item
File "GUI\invisible.py", line 81, in focus_tl_item
IndexError: list index out of range
@Bri🥰

mastodon - Link to source
ahahaha. Yes yes, I'm done making AI slop music. Now it's time for stage 2 of the experiment: We've clearly established Suno's own prompt generator is very much inferior to lyrics of music you yourself write with GPT and spend a bit of time skulpting, reshaping lines, or even prompting GPT a bit more to rewrite concepts and parts of it. That's stage 2 of my experiment: Make more AI music, but this time, spend at least 30-45 minutes building the prompts out for the lyrics and making the lyrics more meaningful through that.
Stage 3 of this will be, make AI music with fully custom-written lyrics, E. AI only used for rhythm and pace of lyrics minimally, not for the writing of the lyrics itself.
The question in all these experiments for me is, "when does AI music go from "slop" to "a useful tool that works to enhance talent?" And how much effort, skulpting is needed for that to happen, could someone pull that off? Suno lets you skulpt instrumentation a bit, but I suspect a key problem to overcome won't be on lyrics alone: The song itself needs to have instrumentation that doesn't make it sound like machines made it alone in composition, and for that I find that Suno doesn't give you many tools.
This entry was edited (2 days ago)

hometown - Link to source
SFFBookClub February Poll :boost_request:

Sensitive content

#sffbookclub

mastodon - Link to source
What's nice about this social media platform is we're like "are you seeing this fucking shit?!" and get responses like "this is the fuckingest shit!" and then some folks go "okay, here are some alternatives to the shit/ways we can try to unfuck some of the shit/resources to slow the impact of the fucking shit" instead of "i don't see what's so bad about the shit" or "i dunno i kinda like the shit" or "woah your paying to much attention too the shit" that you get on the other platforms.
in reply to Tamas G

mastodon - Link to source

miki

Nah. Giving LLMs their identity is done in post-training, and Google can just remove that data for their posttraining mix for Apple. They're definitely going to have one. Since pretraining = ~60-90% of compute, redoing that part won't be that expensive. There are rumors that Google will only ship the base models anyway, and that all the fine tuning will be done internally by Apple (perhaps on Google infrastructure). Those base models have no idea what they are yet.

mastodon - Link to source

RE: mastodon.social/@Gargron/11573…

I have to admit, the performance of a Chromium-based browser compared to Firefox is abysmal. I don't have that many tabs but it lags a lot and the browser constantly puts tabs in hibernation to save memory--something I've never needed with Firefox. So all of you Chrome users live like this? No wonder RAM is in high demand.

Eugen Rochko

2025-12-17 23:24:01

mastodon - Link to source

I've finally switched to the @Vivaldi browser. I've been using Firefox for as long as I've been on the internet, but the focus on AI means it's no longer the browser for me. Thankfully unlike Chrome, Vivaldi supports the uBlock Origin extension which is the most important extension for being able to browse the web nowadays.

mastodon - Link to source

The PAM Duress is a module designed to allow users to generate 'duress' passwords that when used in place of their normal password will execute arbitrary scripts.

This functionality could be used to allow someone pressed to give a password under coercion to provide a password that grants access but in the background runs scripts to clean up sensitive data, close connections to other networks to limit lateral movement, and/or to send off a notification or alert (potentially one with detailed information like location, visible wifi hot-spots, a picture from the camera, a link to a stream from the microphone, etc). You could even spawn a process to remove the pam_duress module so the threat actor won't be able to see if the duress module was available.

github.com/nuvious/pam-duress

#security #Linux #Arch #Debian


GitHub - nuvious/pam-duress: A Pluggable Authentication Module (PAM) which allows the establishment of alternate passwords that can be used to perform actions to clear sensitive data, notify IT/Security staff, close off sensitive network connections, etc

A Pluggable Authentication Module (PAM) which allows the establishment of alternate passwords that can be used to perform actions to clear sensitive data, notify IT/Security staff, close off sensit...
GitHub
#linux #debian #security #arch

Hubert Figuière reshared this.

in reply to diana 🏳️‍⚧️🦋🌱

mastodon - Link to source

miki

One useful addition to this would be a password that does an immediate wipe.

There are situations where that is indeed what you want, while that is extremely conspicuous, if the wipe is irreversible, there isn't much that your captors can do about it afterwards, and it's an option that should be available to users.

For systems that store their disk encryption key in a TPM, you could do this by just destroying the key (though I'm not sure whether PAM would even run in an encrypted disk scenario, I know far too little about how this works on Linux specifically).

mastodon - Link to source
I am getting frothy at all this AI hate. Especially where it comes to music. A good song is a good song, if creativity has been used to create it, adn an AI tool has been used to save thousands on production that most of us don't have, what the hell is the problem? Purists and elitism makes me wild! I don't make a living from my music, but if someone has the brains and the drive to do that, why in heck not!
This entry was edited (3 days ago)
in reply to Marie Elizabeth Bayar

mastodon - Link to source

Andre Louis

I think AI music usage falls into a few different camps.
1. The people that give a random prompt, hit generate and say 'I wrote a thing.' No. No you did not.

2. The people that spend time crafting lyrics and sculpting the AI the way they want, and get amazing results just the way they imagined.
3. The other group of people (me included) who feed it my own music and get it to remix/reboot/reimagine whatever you want to call it, my music in entirely different ways.
Sometimes I'll then take that back to Logic and replay the bits I like and change the bits I don't.

AI in the case of numbers 2 and 3 is a tool then, not a 'be all and end all' bandaid. It should not be that.

The people that hate AI hate it because all the number oners break it for the rest of us, those who care about what we're doing, our craft, our joy, and our care and attention to detail.

That, at least, is my take on it.
Happy to be told otherwise.

in reply to Andre Louis

mastodon - Link to source

Tamas G

What I'm very curious about is where the nuance between that 1 and 2 lies. Like, how much skulpting of the prompt makes it more of your own and not just "generically-written Suno lyrics" - so far, I've only been trying the first, because it's what most people will unfortunately do, but the more generic code-related songs I generate the more it does become clear that while they're catchy, they don't give substance or a story to what the song is meaning to convey, so catchy doesn't always mean useful. And I think that's what the boundary of #1 and #2 are, because if you can bring it above the slop that the AI generated, then yes, it's more than just an AI song. @GamingWithEars @FreakyFwoof @SapphireRose91 @darren_duff @cary5871 @lulu_bear
This entry was edited (2 days ago)
in reply to Michael Marshall

mastodon - Link to source

Tamas G

@TheVoiceGuy yeah, from my little experimenting, you do get better results even if you work with GPT first on shaping a bit of lyrics. Suno's own prompt generator is not great at coming up with exact lyrics, or (like it did with my song about speech synthesizers) it takes things literarly: "softVoice" whispers, and "brailab" lets you see things in the world, because it took the words soft and Braille for their meaning. But then again, it wouldn't even research the meaning of "flexVoice" and "softVoice" because its goal is to give you that prompt in a matter of moments, not to actually create something meaningful. That's what's interesting about it.
@Michael Marshall
in reply to Andre Louis

mastodon - Link to source

Luis Carlos

@Tamasg @SapphireRose91 @darren_duff @cary5871 @FreakyFwoof Good techno! If I may have to do something with it... What people hadn't noticed is that AI music creation is like... Remember George Orwell's 1984? They mentioned a tool to create music... a versificator or something. Are we now using versificators right now? Catch me if I'm wrong
@Marie Elizabeth Bayar @Tamas G @Andre Louis @Darren Duff @Cary ✨

mastodon - Link to source

"Furious AI Users Say Their Prompts Are Being Plagiarized."

Even if we set aside that this is a complaint about plagiarizing the use of a technology that is LITERALLY MADE OF PLAGIARISM, consider:

Say I write up how I made a cool piece of art in some app; if you follow the same steps, do I somehow own your output? And hell, that's not even what this is—it's more like someone posted their Google search terms or sandwich order, and are now mad others are using them.
futurism.com/artificial-intell…


Furious AI Users Say Their Prompts Are Being Plagiarized

A self-described AI ambassador is in shambles after she says other AI prompters have been plagiarizing her instructions.
Joe Wilkins (Futurism)
This entry was edited (2 days ago)

mastodon - Link to source

Consider Amira Zairi, a self-professed “AI educator” and “ambassador” for Adobe, LeonardoAI, and TripoAI, who posted a scathing rant this week on X-formerly-Twitter to her 49,000 followers. Her complaint? Other people were “plagiarizing” her unique AI prompts.

futurism.com/artificial-intell…


Furious AI Users Say Their Prompts Are Being Plagiarized

A self-described AI ambassador is in shambles after she says other AI prompters have been plagiarizing her instructions.
Joe Wilkins (Futurism)

friendica - Link to source
Esquema de lista de que haceres importantes y temas a resolver actuales en y para el establecimiento nuestro, "la tierra de nuestros nietos".

#doDo en #tierraNietos

Tercera semana de verano 2026
05/01/2026 al 11/01/2026desarrollar/mejorar sistema electricidad isla de 12v+20v herramientas portátiles, espacios productivos y mesas de trabajo ***bitPickup*** \* arreglos interior | limpieza parcial de piezas ***IT*** \* desarrollo presencia y publicaciones web

Desarme de placas solares en el techo de quincho.
Armado inicial de placas sobre un palet.
Diseño de uso de enchufes schuco para carga de 12 y 24 voltios.
9/10 y 11 de enero #ciclónExtratropical.

#tierranietos #doDo #ciclónExtratropical
⇧