@miki@dragonscave.space avatar

miki

@miki@dragonscave.space

blind coder / comp-sci student, working in automatic speech recognition for CLARIN. Polish. Libertarian leaning. Feel free to get in touch.

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

miki, to random
@miki@dragonscave.space avatar

Google Drive for Mac went from hands-down the best cloud app I've ever seen on any platform to an utter piece of garbage.

miki,
@miki@dragonscave.space avatar

@SwordArtStreamer Not as bad as Google Drive on Windows, but pretty bad. Folders take a ton of time to load, even if they're not too big and when on a good connection, files have to be downloaded in their entirety before they can be opened (this didn't use to be the case), copying to disk is slow, there's no progress bar when copying, attributes are copied over (if you download a folder you don't have write rights to, you can't delete that folder from your own disk without entering your Mac password), QuickLook doesn't work and so on.

Caoimhe, to random
@Caoimhe@dragonscave.space avatar

I'm going to spend the next 2 days in a car, so I need any recommendations from you for books, music, movies, or anything you like that I can watch/read/listen to so I don't go crazy out of boredom. Feel free to share your favorite pieces of media that you feel like sharing to save me.

miki,
@miki@dragonscave.space avatar

@x0 @Caoimhe There's also fichub.net (which is less hassle to use and works on mobile, if you want to download the next part of something if you're on the go).

ftdl, to random Polish
@ftdl@pol.social avatar

Komunikat NapiGen

Dzień dobry. W ten weekend dział infrastruktury FTdL rozpoczął uruchamianie nowego serwera dedykowanego pod rozwiązania LLM w naszej Krakowskiej serwerowni KRK-DC.

Mogą wystąpić przejściowe problemy w działaniu "starego" Napigena. O uruchomieniu nowego i zakończeniu migracji poinformujemy osobnym komunikatem oraz wiadomością e-mail do osób korzystających z niego.

Trwają również prace nad stroną www projektu oraz nad otworzeniem go jako ogólnodostępny projekt FOSS z kodem i materiałami na Codeberg.

Przypominamy, że istnieje możliwość uczestnictwa w programach Early Access oraz Ambasadorskim projektu.

Możecie również dołożyć swoją cegiełkę do rozwoju projektu, wspierając zbiórkę na sprzęt: potrzebujemy dodatkowych CPU, RAM, GPU oraz dysków na storage.

Zbiórka tutaj -> <https://ftdl.pl/sprzet-llm-napigen/>

miki,
@miki@dragonscave.space avatar

@ftdl Nie myśleliście o zaangażowaniu społeczności, pozwoleniu użytkownikom na udostępnianie własnych GPU na rzecz projektu, na modłę fold@home czy seti@home?

Pracowałem nad takim projektem jako koncept na studiach (oparte to było na WHisperze), powinienem jeszcze mieć gdzieś do tego kod. Trzeba by dorobić jakieś uwieżytelnianie, ratę limity i crosswalidacje między użytkownikami (w celu zapobiegania hakowaniu systemu i wsadzaniu obraźliwych treści), ale myślę, że idea jest fajna.

ZBennoui, to random

So over the last several months, I've been looking at all of these AI generated music services. Suno, Udio, and now the new one from Eleven. As someone who generally has a pretty positive outlook on machine learning/AI, I think these tools are really interesting and a great way to help people who don't have any musical ability make personalized tracks. However, I take issue with how these systems are trained. It's pretty much been confirmed that Udio is trained on vast amounts of copyrighted material, very likely without consent considering how new the company is. With Suno it's hard for me to tell, but others have theorized that there's copyrighted stuff in there as well. These companies are telling you that you are allowed to use whatever you generate for commercial purposes, but I fail to see how they have the right to do so. I wonder if artists are even aware that their songs could potentially be included in these models, and honestly just the whole ethos of these companies is disgusting. What gives them the right to scrape massive amounts of copyrighted material these people spent a crazy amount of time on, just to dump it into a model that can generate whatever you want based on a simple text prompt? Let me be clear that I have no problem with the Technology itself, I think it's really cool, but the only reason it's able to sound as good as it does is because they are training it on a lot of music that they don't have the rights for. Take a look at Stable Audio if you want to see whats possible with just licensed royalty-free tracks, spoiler it's nowhere near as good. Some of that could of course be due to the architecture, but more likely it's the data they had access to while training. I wonder what Eleven used to train their models, but considering how clean the results are, I suspect they got custom multi-tracks from whoever they decided to work with. I'm personally far more excited about what Apple is doing in this space with the new session players in the upcoming Logic updates, and I hope this will be the path forward rather than massive audio generation models trained on unlicensed material.

miki,
@miki@dragonscave.space avatar

@ZBennoui The reasoning here is that actual musicians are also trained on copyrighted content, and yet as long as the content they create isn't too close to an already existing song, they're fine.

Think how bad a modern musician would be if they were only allowed to listen to royalty-free music.

miki,
@miki@dragonscave.space avatar

@ZBennoui We genuinely don't know what the courts are going to decide (and it's probably going to both differ between jurisdictions and be superseded by new laws in some places). It's extremely unlikely for AIs to be outlawed in every single country after those court battles are over. Anybody who says otherwise (regardless of what side they're on) is just trying to push their narrative IMO.

This is genuinely new technology, different from anything we've seen before, and there are no easy answers here.

miki,
@miki@dragonscave.space avatar

@ZBennoui Then there's the politics, even if US courts outlaw AI training but China doesn't, the US will have very little choice but to amend their laws and catch up.

miki,
@miki@dragonscave.space avatar

@ZBennoui I really don't have much of a problem with this, just as I don't have a progblem with programmers looking at and learning from my code. People don't seem to be able to grasp the difference between learning from something and straight up copying it.

miki,
@miki@dragonscave.space avatar

@ZBennoui As long as the model doesn't do this unintentionally, this should be on the person prompting the model to do this, not the model makers.

If you buy a CPU from Intel and make it execute the instructions of a malicious program you wrote, you're responsible for the harm caused, not Intel.

miki,
@miki@dragonscave.space avatar

@ZBennoui The way I think about this is that you don't punish musicians for being able to perform Beatles songs when forced to do so at gunpoint, and so you shouldn't punish models for performing these songs when cleverly prompt-engineered. Instead, you should punish the promt engineer themselves, just like you'd have an issue with the person holding the gun. Now when models do this unintentionally, which sometimes (although rarely) happens, that's a far muddier issue.

miki, to random
@miki@dragonscave.space avatar

Is this just me, or are university websites disappearing?

It feels like in the 2000's, quite a few professors (both in the US and otherwise) would publish their notes and lecture slides on their website, usually under a university domain. Those materials were often of extremely high quality, as they were published by experts in the field, with no expectation of profit. I've been noticing that a lot of links to those old materials that one can find on the web don't work any more, and a lot of new content is uploaded to internal learning management systems instead. Is this just my impression or is this actually true?

mekkaokereke, to random
@mekkaokereke@hachyderm.io avatar

If you got upset at me for pointing out that more people died of Covid under Biden than Trump, in large part because Biden rolled back common sense restrictions that were in place under Trump...

Then you'll hate me pointing out that Trump introduced a 25% tariff on Chinese EVs, and Biden is upping that to 100% tariff.

https://insideevs.com/news/719283/chinese-ev-tariffs-biden-quadruple/

We can't allow the US to get off of fossil fuels... unless US billionaires win! 🤡

Fear the BYD Dolphin! (an EV car for $12K)
https://m.youtube.com/shorts/T3nfyO_UHjk

miki,
@miki@dragonscave.space avatar

@mekkaokereke Genuine question, I don't know much about this industry. Are chinese EVs cheap becauce the Chinese know how to make them cheap, or are they cheap because of government subsidies, exploiting child labor etc?

danjac, to random
@danjac@masto.ai avatar

It's downright criminal how universities and other courses encourage or force students to set up private AWS accounts.

AWS is for companies, period. It should never be used by a private individual. If it were designed for private individuals, they would have implemented hard limits on the free tier rather than their laughable billing alerts.

You should never use it for coursework or side projects: the danger is too high.

https://old.reddit.com/r/aws/comments/qgr9jh/was_billed_60k_with_a_free_tier/

miki,
@miki@dragonscave.space avatar

@danjac Azure has a student tier for exactly that use case. There's no credit card required, the worst you can do is accidentally empty your account and be unable to participate in the rest of the course, and even that can be remedied by the professor somehow I think.

miki, to random
@miki@dragonscave.space avatar

How unoriginal do you have to be to name a new city after an already existing place?

osma, to telegram
@osma@mas.to avatar

, May 8 assessment: "Reports indicate that there is an available open-source tool that allows people to search by specific coordinates for Telegram users who have enabled a certain location-sharing setting."

But of course there is. The russian-engineering, roll-your-own-crypto, cryptocoin-shilling, encrypted-but-not-encrypted messaging app to have a zero-day exploited privacy flaw exposing users' location? I can't imagine where such failures would come from.

miki,
@miki@dragonscave.space avatar

@osma Why is this news? Of course a "find people nearby" feature lets you find people nearby, and it should be obvious to anybody with even a passing knowledge of cybersecurity that it's impossible for any app to protect itself against malicious users spoofing their location. If you're on Android, you don't even need a special tool for that, there's a way to set a gps-spoofing app right in developer settings.

miki, to random
@miki@dragonscave.space avatar

I think we severely underappreciate the impact of IP address exhaustion on basically everything on the modern internet.

If not for that issue, dynamic IP addresses and NATs wouldn't have to exist, and that would make P2P-based protocols a lot more practical.

miki, to random
@miki@dragonscave.space avatar

Honestly, the main argument against training LLMs on fedi content isn't ethics or consent issues, it's is extreme bias and lack of diversity. Training on Mastodon might be just as bad as training on 4chan or Tumblr in some ways.

miki, to random
@miki@dragonscave.space avatar

Google PageRank is like SMTP OpenRelay.

A really nice idea that makes life easier on a small-ish network where most participants can be trusted, absolutely terrible on a network whose size approaches infinity and anybody can publish anything anonymously.

albertcardona, to Trains
@albertcardona@mathstodon.xyz avatar

In Europe, flying is cheaper than taking the train.

It's an embarrassment, and a major problem: we have to stop flying for silly short distances. Realise that the overheads of flying (reaching the airport, awaiting 2 hours, the flight, the unloading, reaching the destination) largely cancel out any time gains of flying. And the carbon costs are utterly untenable. Not to speak of the modern, dire conditions of the whole flying "experience".

Another embarrassment is that train connections can't be guaranteed when across countries or companies. They aren't even coordinated. As if those who commission and set the schedules didn't travel by train themselves, at least not internationally. In considering how tiny most European countries are, it's frankly bizarre.

There are so many destinations one could travel by train to, yet in practice, it's not sensible. A disgrace.

The upside is that it can be fixed.

miki,
@miki@dragonscave.space avatar

@albertcardona The unsaid part of this is that the cheap airlines are private (and hence efficient), while most train companies are either public or semi-public, and hence both expensive and inefficient. If you take a look at Polish train prices, which aren't that bad compared to other places, going by car instead becomes more affordable at 3 to 4 people, assuming normal tickets. This makes no sense, there's no way a car is more efficient than a train.

miki,
@miki@dragonscave.space avatar

@cybertailor @albertcardona Do the cheap airlines get any of these? I was under the impression that most subsidies go to the expensive state-owned enterprises like Lufthansa, KLM, British Airways or Lot.

miki, to random
@miki@dragonscave.space avatar

It took Python over 30 years to make "exit" in the terminal actually exit instead of telling you what to do to exit.

I know why this works the way it works (the message is actually the string representation of the exit function, and making the string representation of "exit" quit the interpreter would be just as counterintuitive), but this ugly hack should have been fixed in like 0.2 for God's sake.

miki,
@miki@dragonscave.space avatar

@Caoimhe They finally fixed it in 3.13, which is in Beta.

miki,
@miki@dragonscave.space avatar

@Caoimhe @x0 These days, there's very little reason to actually use Python on Windows though. Even if that's what you're on, you probably should be using WSL for most things instead.

miki, to random
@miki@dragonscave.space avatar

There's a fundamental conflict of interest between accessibility experts and accessibility overlay vendors. If accessibility overlays ever get good enough to replace the experts, which, just to be clear, they haven't yet, they'll put the experts out of their jobs. I wouldn't be surprised if accessibility "experts" keep spreading FUD about overlays long past the point when they actually become useful. After all, making themselves sound irreplaceable might be the only viable strategy for them.

This is just something to ponder when hearing what experts say about AI in accessibility.

matt, to random

Context on my last boost (https://dragonscave.space/@jscholes/112406794436648906) for my followers who aren't plugged into the blind community: Sonos recently released an update to their mobile app that makes it way less usable with VoiceOver on iOS, and presumably TalkBack on Android as well. @jscholes has insightfully connected this to the genericization of UI development, and this post in particular: https://www.baldurbjarnason.com/2024/react-electron-llms-labour-arbitrage/

miki,
@miki@dragonscave.space avatar

@matt I'm getting more and more convinced that slapping an AI-based accessibility overlay is eventually going to be the solution, and it's going to happen basically before anything else. We wont turn around the screen-reading world in 5 years, and I wouldn't be surprised if we got working overlays by then.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • normalnudes
  • tsrsr
  • DreamBathrooms
  • thenastyranch
  • magazineikmin
  • khanakhh
  • hgfsjryuu7
  • Youngstown
  • slotface
  • vwfavf
  • rosin
  • ngwrru68w68
  • kavyap
  • PowerRangers
  • Leos
  • ethstaker
  • cubers
  • everett
  • modclub
  • InstantRegret
  • tacticalgear
  • Durango
  • mdbf
  • cisconetworking
  • tester
  • GTA5RPClips
  • osvaldo12
  • anitta
  • All magazines
    •