riskable

AI models aren't trained on anything "stolen". When you steal something, the original owner doesn't have it anymore. That's not being pedantic, it's the truth.

Also, if you actually understand how AI training works, you wouldn't even use this sort of analogy in the first place. It's so wrong it's like describing a Flintstones car and saying that's how automobiles work.

Let's say you wrote a book and I used it as part of my AI model (LLM) training set. As my code processes your novel, token-by-token (not word-by-word!), it'll increase or decrease a floating point value by something like 0.001. That's it. That's all that's happening.

To a layman, that makes no sense whatever but it's the truth. How can a huge list of floating point values be used to generate semi-intelligent text? That's the actually really fucking complicated part.

Before you can even use a model you need to tokenize the prompt and then perform an inference step which then gets processed a zillion ways before that .safetensors file (which is the AI model) gets used at all.

When an AI model is outputting text, it's using a random number generator in conjunction with a word prediction algorithm that's based on the floating point values inside the model. It doesn't even "copy" anything. It's literally built upon the back of an RNG!

If an LLM successfully copies something via it's model that is just random chance. The more copies of something that went into its training, the higher the chance of it happening (and that's considered a bug, not a feature).

There's also a problem that can occur on the opposite end: When a single set of tokens gets associated with just one tiny bit of the training set. That's how you can get it to output the same thing relatively consistently when given the same prompt (associated with that set of tokens). This is also considered a bug and AI researchers are always trying to find ways to prevent this sort of thing from happening.

No it can't do that. It's an LLM, it can only generate the next word in a sequence.

Your knowledge is out of date, friend. These days you can configure an LLM to run tools like curl, nmap, ping, or even write then execute shell scripts and Python (though, in a sandbox for security).

Some tools that help you manage the models are preconfigured to make it easy for them to search the web on your behalf. I wouldn't be surprised if there's a whole ecosystem of AI tools just for searching the web that will emerge soon.

What Mozilla is implementing in Firefox will likely start with cloud-based services but eventually it'll just be using local models, running on your PC. Then all those specialized AI search tools will become less popular as Firefox's built-in features end up being "good enough".

Have you tried using an LLM configured to search the Internet for you? It's amazing!

Normal search: Loads of useless results, ads, links that are hidden ads, scams, and maybe on like the 3rd page you'll find what you're looking for.

AI search: It makes calls out to Google and DDG (or any other search engines you want) simultaneously, checks the content on each page to verify relevancy, then returns a list of URLs that are precisely what you want with summaries of each that it just generated on the fly (meaning: They're up to date).

You can even do advanced stuff like, "find me ten songs on YouTube related to breakups and use this other site to convert those URLs to .ogg files and put them in my downloads folder."

Local, FOSS AI running on your own damned PC is fucking awesome. I seriously don't understand all the hate. It's the technology everyone's always wanted and it gets better every day.

Did you change the ink during this process? I've found things inside printers before.

It's called dogfooding and it's what you're supposed to do to improve your product.

Total market share is irrelevant. What matters more is total users.

If you make a product and there's a million people on a platform who could buy it, the costs to port that product (and support it) need to be low for it to be worthwhile.

If the total number of people on that platform increases to 10 million, now the cost to port/support becomes more like a minuscule expense rather than a difficult decision.

When you reach 100 million there's no excuse. There's a lot of money to be made!

For reference, the current estimated amount of desktop Linux users globally is somewhere between 60-80 million. In English-speaking countries, the total is around 19-20 million.

It's actually a lot more complicated than this, but you get the general idea: There's a threshold where any given software company (including games) is throwing money away by not supporting Linux.

Also keep in mind that even if Linux had 50% market share, globally, Tim Sweeney would still not allow Epic to support it. I bet he'd rather start selling their own consoles that run Windows instead!

One thing for certain, Microsoft will not stop using Copilot to develop their software in house.

You're wrong, but I think you'll be OK with that because the reality of the situation is actually hilarious:

https://www.theverge.com/tech/865689/microsoft-claude-code-anthropic-partnership-notepad

"Turns out Copilot sucks so let's just use our competitor's superior product but that's no reason we can't keep foisting the inferior garbage on the masses!"

Random chance gambling. All throughout history there have been many statistical anomalies and if I joined them in such a competition, I'd have just as good a chance at winning as anyone else.

That's just the tip of the iceberg of cool and useful stuff you can do with KDE Plasma (and Kwin).

Another tip: Did you know that KRunner (Alt-Space) can do unit conversions? Type Alt-Space and 10cm or something like that and it'll give you that value in inches.

Another: You can bind shortcuts to mouse buttons like Ctrl-Alt-Right (click) And Ctrl-Alt-Left to say, switch desktops right/left.

You can type Ctrl-i in Dolphin to filter files. So if you're looking at your enormous downloads directory and you want to see all the .png files you can type Ctrl-i, png and it'll only show you files with png in their name.

KDE's "get hot new stuff" framework works with Dolphin "actions" (context menu file handlers) so you can go into the settings—>Context Menu and click on "Download New Services" to browse tons of free scripts/tools that let you do things like file conversions, write disk images to USB drives, get checksums, etc.

I actually made a personal script that converts videos to looping .webp files (or just sets WebP files to loop forever). So I can right click on a .WebP, .webm, .mp4, etc and it'll run ffmpeg on it in the background.

Every decade since 1999 (the year of the Linux desktop—for me) I spend a few weeks trying out all the hot new shit in terms of desktop environments. I'll switch to Gnome for a few days, get disappointed at how much I miss from KDE, and then try one of the newer ones like Cosmic. Then I'll play with the latest versions of the classics (xfce) and marvel that they still make you configure everything in a single file or they still lack basic shit that normal people want like a clipboard manager.

All the actually useful or just plain really, really nice/handy stuff is built into KDE Plasma. I've been using so many of those features for so long, I can't fathom having to go back to a world without say, being able to navigate the filesystems on all my other PCs via ssh:// (and other KIO workers).

I remember when KDE 2.0 came out and it added support for kioslaves (now called KIO Workers) and it completely changed how I viewed desktops. That was in the year 2000. How is it that literally nothing else (not other FOSS desktops nor Windows or Macs) has implemented the same feature?

It's not just the file manager, either. I can access ssh:// (or any other KIO worker) from any file dialog! The closest thing is shared drives in Windows but even that isn't nearly as flexible or feature rich (or efficient, haha).

Then there's the clipboard manager (klipper), Activities, and a control panel that lets you customize everything to extreme degrees. It even supports fractional scaling and has supported that since forever. I remember when they introduced that feature over a decade ago and it still blows my mind to this day just how forward thinking the devs were.

Monitors since forever have had a different X DPI than the Y DPI. Yet only the KDE devs bothered to both query the monitor's DDC info to figure that out and set it correctly when the desktop starts.

There's other features that drive me nuts when I don't have them! For example, the ability to disable global shortcuts on specific windows. So if I've got a remote desktop open to my work I can send Super-. (Win-.) and that'll open the Windows emoji picker in the remote desktop instead of the KDE one (locally). And it will remember this setting for that application!

I can make any window I want stay above others temporarily to take notes, enter values into the calculator, or just turn any window into something like a HUD (you can control any window's transparency on the fly!).

It even supports window tiling! A feature most people aren't aware of. Like, if you're already running KDE, why bother with a tiling window manager? You've already got it (though the keyboard shortcuts to manage the tiling layout in real time are lacking).

TL;DR: KDE Plasma is the best desktop in existence across all platforms and this is easy to prove with empircal evidence.

Everyone that said they "dropped a bomb in the toilet" is just a poser. This guy is the real deal.

It's no different than other kinds of voting: It only works if everyone is well-informed.