Welcome to Incremental Social! Learn more about this project here!
Check out lemmyverse to find more communities to join from here!

@ayaya@lemdro.id cover
@ayaya@lemdro.id avatar

ayaya

@ayaya@lemdro.id

This profile is from a federated server and may be incomplete. Browse more on the original instance.

ayaya ,
@ayaya@lemdro.id avatar

What you're asking for is literally impossible.

A neural network is basically nothing more than a set of weights. If one word makes a weight go up by 0.0001 and then another word makes it go down by 0.0001, and you do that billions of times for billions of weights, how do you determine what in the data created those weights? Every single thing that's in the training data had some kind of effect on everything else.

It's like combining billions of buckets of water together in a pool and then taking out 1 cup from that and trying to figure out which buckets contributed to that cup. It doesn't make any sense.

ayaya ,
@ayaya@lemdro.id avatar

Sorry, I misinterpreted what you meant. You said "any AI models" so I thought you were talking about the model itself should somehow know where the data came from. Obviously the companies training the models can catalog their data sources.

But besides that, if you work on AI you should know better than anyone that removing training data is counter to the goal of fixing overfitting. You need more data to make the model more generalized. All you'd be doing is making it more likely to reproduce existing material because it has less to work off of. That's worse for everyone.

ayaya ,
@ayaya@lemdro.id avatar

If the model isn't overfitted it's also not even copying. By their nature LLMs are transformative which is the whole point of fair use.

ayaya ,
@ayaya@lemdro.id avatar

Again, even an exact copy is not stealing. It's copyright infringement. Theft is a different crime.

But paraphrasing is not copyright infringement either. It's no different than Wikipedia having a synopsis for every single episode of a TV series. Telling someone about what a work contains for informational purposes is perfectly fine.

ayaya ,
@ayaya@lemdro.id avatar

Same here. Switched to Arch in 2015 so I am also coming up on the 9 year mark. I have had very few issues, and the ones I have had were usually my fault for doing something stupid. I used Windows, OS X, and Ubuntu previously and compared to those Arch is a dream. Hence why I've stuck with it for so long now.

ayaya ,
@ayaya@lemdro.id avatar

https://librewolf.net/

ayaya ,
@ayaya@lemdro.id avatar

Obviously you've never used Arch btw. We live for the sudo pacman -Syu.

ayaya ,
@ayaya@lemdro.id avatar

$80? I run mine on a Pi Zero that I got for $9 with a $6 wired network adapter for a grand total of $15. No problems for a household of five with one of us (me) being an extremely heavy user.

ayaya ,
@ayaya@lemdro.id avatar

I used to do that, but it comes with the problem of your DNS going down any time you want to restart or do a hardware swap on your NAS. Or since it was running in docker something as simple as reloading docker would knock out the internet for a few minutes. It's worth the $15 to have them operate separately.

ayaya ,
@ayaya@lemdro.id avatar

I pretty much never reboot the Pi. It currently has over 18 months of uptime on it. My NAS on the other hand I probably restart for one reason or another maybe once every 6 months. So yeah I'd say I reboot it minimum 3x more often.

Plus a reboot takes much longer on my NAS than on the Pi. The server board is slow to start, the SAS cards are slow to start, and unRAID is slow to start. Then I need to manually enter the password for disk encryption. Then wait for the array to start up. Then wait a bit more for the docker containers to start. Add all of that up and even the absolute fastest reboot is like 10 minutes while the Pi probably takes 30 seconds.

And what if I want to swap hard drives? Now it's down for an hour. I guess I could wait until 3am to do all my upgrades so everyone is asleep, but I'd rather not. I suppose if it were just for myself it would matter a lot less. But again, it's only $15 to not have to think about it at all.

ayaya ,
@ayaya@lemdro.id avatar

It works in Elden Ring and Baldur's Gate 3 for me. Haven't tried anything else.

ayaya ,
@ayaya@lemdro.id avatar

Yeah there are like 5 monitors with full array local dimming, most being $500+ except for that one AOC. And OLEDs are still $700+ and have burn-in after a year of desktop use.

ayaya ,
@ayaya@lemdro.id avatar

Yep I don't even play that many games but I watch a lot of movies/TV. HDR works great in mpv. Couple of tweaks in your mpv.conf and you're off to the races.

ayaya , (edited )
@ayaya@lemdro.id avatar

Yes it's an exaggeration but it's not far off. The one for $290 is the aforementioned AOC.

This isn't a perfect list but pcpartpicker only has 15 monitors with HDR1000 or higher with one being a duplicate so it's actually 14. If you remove the HDR filters there's 773 monitors.

That means only 14 out of 773 monitors support HDR properly. And that doesn't even mean they're good, just that they support it.

And oops I should have specified 27 inches or under, that is my bad. 27 inches is what I was shopping for recently. Personally I actually prefer 24 inches but they pretty much stopped making good 24 inch ones.

ayaya ,
@ayaya@lemdro.id avatar

I work from home and never call anyone so on Tello I pay $6/mo for 100 minutes + 1GB of data that pretty much functions as a 2FA delivery system.

ayaya ,
@ayaya@lemdro.id avatar

I've never had a job where we celebrate Shitty Ass Day. What's that like?

ayaya ,
@ayaya@lemdro.id avatar

You are looking for S0und

Google's "Manifest V2" Chrome extension phaseout next month is expected to impact the original uBlock Origin extension, which still uses the V2 framework and has 37 million users (www.theregister.com)

The new MV3 architecture reflects Google's avowed desire to make browser extensions more performant, private, and secure. But the internet giant's attempt to do so has been bitterly contested by makers of privacy-protecting and content-blocking extensions, who have argued that the Chocolate Factory's new software architecture...

ayaya , (edited )
@ayaya@lemdro.id avatar

Same story for me on a OnePlus 5T which is the even older Snapdragon 835. Firefox is genuinely unusable. I tried Mull and Iceraven too. For several months I tried to put up with it, but they were all a slow and buggy mess. Switched to Brave and it works fine.

I use Librewolf on my desktop for the record.

ayaya ,
@ayaya@lemdro.id avatar

It is genuinely amazing, there's a reason us Arch users never shut up about it. The setup/configuration in the beginning will seem daunting but once you have everything the way you want it is a smooth and enjoyable experience.

ayaya ,
@ayaya@lemdro.id avatar

If that's the case it should be a much easier transition. I also came from Ubuntu originally.

You will notice the difference right away. Everything is always up to date so you're not waiting half a year for updates and there's no big upgrade transitions between major versions. And pacman is a lot faster than apt in general.

Plus with the AUR you'll never touch another PPA again. Almost anything you could possibly want is in there, even some of the most obscure/specific things.

I do recommend doing everything from scratch if you have the time, but if not EndeavourOS is literally just preconfigured Arch (and I do mean literally, it uses the same repos) so that's also a solid option.

ayaya ,
@ayaya@lemdro.id avatar

It brings nothing new to the mix until you want something that's up to date or something that's not in the main repos and now you have to track down a PPA or manually install a deb file and keep it updated yourself instead of being able to use the package manager.

ayaya ,
@ayaya@lemdro.id avatar

There is also FreshTomato if your router has Broadcom wifi chipset like mine does.

ayaya ,
@ayaya@lemdro.id avatar

I definitely got stuck on their levels in Megaman as a kid.

ayaya ,
@ayaya@lemdro.id avatar

Yeah it can certainly cause problems, it's just not ADHD.

ADHD doesn't even really mean short attention spans, it's more of the inability to willingly direct attention. It's the same way people incorrectly use "OCD" to mean liking things clean and/or orderly.

I have ADHD and I've had times where I've done the same thing for 14 hours straight (even forgetting to eat) when my brain decides it wants to latch onto that thing. You just need to be sufficiently stimulated, hence why stimulants can work as a treatment.

ayaya ,
@ayaya@lemdro.id avatar

Exactly. If all you want to do is torrent then it's by far the best option. $2.22/mo ($80 for 3 years) which is less than half the price of anything else, has portforwarding, and with wireguard I can saturate a full gigabit no problem on private trackers.

ayaya ,
@ayaya@lemdro.id avatar

It was also already in Arch's KDE-unstable repo. I've been using Plasma 6 for like 3 months.

Affordable Android Excellence: Best Smartphones Under $200 in 2024 (www.gizchina.com)

While flagship smartphones boast impressive features, spending $1,000 is not a prerequisite for a satisfying Android experience nowadays. If you’re in need of a new smartphone and have a budget of approximately $200, there are numerous excellent options available. Surprisingly, some of the best Android phones under $200 come...

ayaya ,
@ayaya@lemdro.id avatar

It doesn't necessarily defeat the point if the only reason you are using Lineage is for OS updates and not for privacy reasons. That was my original reason for using it before de-googling.

I don't have google play services anymore but I do still use microG just for Revanced because I am a psychopath that actually likes YouTube recommendations.

ayaya ,
@ayaya@lemdro.id avatar

Especially for hard drives. 8TB SAS drives are down to about $45 a piece.

Brand new enterprise-grade 8TB drives are more around $180 new. Meaning as long as you have redundancy (which you should anyway) then you can lose four used drives before it stops being worth it. Not to mention drives get cheaper so if your $45 drive dies 2 years from now you could probably replace it for $35 etc.

ayaya , (edited )
@ayaya@lemdro.id avatar

Just buy them on eBay. Why does it matter where they come from? Again, four of them have to die before it's no longer worth it. It's extremely unlikely you'd be that unlucky.

Personally I have 15 drives in my NAS, all of them were bought used and they've been running 24/7 for 4+ years without issue. Originally I expected to lose at least one per year but they just keep chugging along. All of them have at least 40k power on hours, with the oldest 3TB ones having over 80k (9+ years)

I use unRAID so if/when one does die it's as simple as pulling out the dead one, popping in a new one, and letting it rebuild itself.

ayaya ,
@ayaya@lemdro.id avatar

I recommend reading the actual paper this article is about. DDG is actually by far the worst by their measures. Google is 9% spam compared to 31% for DDG and 23% for Bing. That's a huge difference.

I would recommend trying a SearXNG instance if you haven't before. You can combine results from multiple sources. I use Google as my main source while also having access to the DDG-style !bangs.

ayaya ,
@ayaya@lemdro.id avatar

I recommend reading the actual paper. DDG is actually by far the worst by their measures. Even though Google has gotten worse, it's results are still massively superior to the competition. 9% spam compared to 31% for DDG and 23% for Bing. That's a huge difference.

I would recommend trying a SearXNG instance if you haven't before. You can combine results from multiple sources. I use Google as my main source while also having access to the DDG-style !bangs.

ayaya ,
@ayaya@lemdro.id avatar

You mean Vanguard, which was announced but isn't actually in the game yet. Their plan is to add it late February or early March. We don't actually know any details about the implementation except that it won't be used in the macOS version.

This new data poisoning tool lets artists fight back against generative AI (www.technologyreview.com)

A new tool lets artists add invisible changes to the pixels in their art before they upload it online so that if it’s scraped into an AI training set, it can cause the resulting model to break in chaotic and unpredictable ways....

ayaya ,
@ayaya@lemdro.id avatar

Obviously this is using some bug and/or weakness in the existing training process, so couldn't they just patch the mechanism being exploited?

Or at the very least you could take a bunch of images, purposely poison them, and now you have a set of poisoned images and their non-poisoned counterparts allowing you to train another model to undo it.

Sure you've set up a speedbump but this is hardly a solution.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • random
  • incremental_games
  • meta
  • All magazines
    •