Offline-only speech-to-text, integrated with the desktop for push-to-talk voice typing? That's the kind of AI that I'd like to see. Actually add features that can help people without harming their rights. I'm still moving new machines to Debian but this is nice.
Also how is speech to text AI? It has existed for decades, obviously a lot better now but I don't think I'd consider it "AI"
There's been ML and non-ML ways of doing STT over the years. as far as I recall. The current best implementations are ML-based. In coloquial terms ML algorithms are AI. We used to call them AI in the 2010s, before AI was (un)cool.
And 'AI' has existed since the 1950s.
I have been using a speech-to-text AI system the last day, and I'm using a ~~whisper large 3 turbo~~ Voxstral Small (Mistral) and a rewording model that fixes the sentence but doesn't rewrite it, and it's almost perfect. I'm using European-hosted AI through cortecs.ai, and it's really cheap.
Being real, this is why I fucking hate the bullshit, corporate greed hype of LLMs and generative software. All the "bubble" shit? It tars all versions of the technology with the same brush.
This? This is exactly what it should be used for. And, ffs, earlier speech to text was really the same fucking thing in essence. Software that took input in the form of voice, compared it to a set of data, and made a best guess at what you meant. Yeah, the details are different, but it's the same concept.
This? This is fucking awesome. Locally run, and doing a job that's vital in accessibility, with the side benefit of being useful to others. Assuming canonical is being honest anyway.
But this kind of thing should be the way things are done.
That's why we should not call everything AI
But it is AI, so it should be called that. People should adjust their simplistic notions about the term instead
Nah, it isn't. Intelligence implies independence. What it is is a fancy algorithm with a big data set.
It doesn't have to be general ai to be called ai, but so far none of the models I'm aware of have reached a standard to be called intelligence in the colloquial sense for sure
The term used in academic literature and the field itself for that kind of technology is and has been AI for at least ten years. Intelligence doesn't imply independence anyways? And besides, even if it did, thats why an entire 50% of the term consists of the modifier "artificial". So like, you're right that it isn't intelligence in the colloquial sense. It's artificial intelligence in the technical and standardized sense, though. The use of term is pretty much totally undebatable. Just because people don't like the term now doesn't change that.
Even if we called them LLMs, which we should, people will keep having these negative connotations to the technology because of overmarketing. This feature is still using LLMs, and that's not supposed to be a bad thing.
Stop blaming the technology, and blame the corpos pushing it to the moon. This is "BitTorrent is only used for piracy" bullshit all over again.
This would have been better received if they just didn't use AI in the name. Sure, it's just using an LLM under the hood, but it's running purely locally. It also betters Linux since it helps address an accessibility issue.
Implicit optional features to use local LLMs for STT is something that I think most reasonable people could get behind. Too many accessibility tools for the disabled sit behind paywalls and subscription models.
Bit of a click baity title lol. If there’s one good use of AI, its probably accessibility.
Yeah, "Ubuntu wants you to use their new feature" is... unsurprising. Explaining the benefits and purpose of that feature? Now you're talking.
Well they can listen to my voice when I say this:
The framework split things into two groups, implicit AI that quietly improves what you already use and explicit AI that are features you'd actually summon on purpose.
The very first paragraph already upsets me. Have in mind, I would criticize this on every other operating system too. I believe no one should use Ai tools that act autonomously in the background, to improve or change what you already use. It should always be a "summon on purpose".
Canonical can take its AI and walk into the sea
You don't like accessibility?
Not for that price.
The "price" of a free offline speech to text AI model? Three of them, actually, to work with varying levels of compute resources available?
You anti-AI folks are friggin' ridiculous.
I don’t think it’s anti-AI more a lack of trust of services saying here’s a product…and the concern it will be used for ulterior motives. I know I don’t like my voice being captured.
It's captured when you press a button and only handled locally. This is exactly the sort of thing you want for accessibility. Not everybody can type or type well.
What do you think are the ulterior motives behind this GPLv3 licensed app? I think the ulterior motive is getting better accessibility tools so it's easier to convince governments to use Ubuntu (and likely pay for extended support).
Fuck AI.
Well "reasoned" argument there. You'll go far in MAGA.
You want "reason"? AI is a jumped up chatbot with delusions of grandeur that jacks up energy prices, consumes a stupid amount of scare water and throttles the supply of computer parts just to tell people there are 5 Rs in strawberry and that they should kill themselves. Not to mention the outright theft of the works of authors, designers, and artists.
FUCK. AI.
And fuck you for comparing me to those shit dribbling twats in MAGA.
Nothing AI is free. Unless there's a chain of custody for all of the training data, it's still unethical even if it's used for a good thing. If I build a wheelchair ramp out of the flesh and bones of orphans I'm still not a very good person. And there are non-AI ways to accomplish this that are just as good that would require almost comically less resources.
This attitude is why Ubuntu, and only Ubuntu, recommends a minimum of 6 GB of Ram btw. You can run a full KDE system with onboard graphics and all the bells and whistles for less than 2 GB on other distros.
If I build a wheelchair ramp out of the flesh and bones of orphans I’m still not a very good person.
Comparing copyright, a bullshit concept that was designed only to protect the rich, to building with "the flesh and bones of orphans" is so extreme that it's not even worth arguing about.
Nothing AI is free. Unless there’s a chain of custody for all of the training data, it’s still unethical even if it’s used for a good thing.
This is the weirdest sort of AI bullshit I keep coming across.
And there are non-AI ways to accomplish this that are just as good that would require almost comically less resources.
And... Where are they?
I think Ubuntu is the Windows 11 of Linux distros.
Thanks Canonical.....I'll just throw it in the pile with all the other "wonderful" things you've made. It can go on the shelf next to Mir.
Erm, no.
Come on.
Offline-only is privacy-respecting. Accessibility is a noble goal.
All in all, if there's an AI usecase that's as morally acceptable as it gets, it's this one.
I get that it's Ubuntu of all people, but even Big Tech produces some ideas every now and then that FOSS lovers can get behind and democratize!
God fuck them. They are hunting for hype and not making linux better for the masses
Out of curiosity would it be accurate to call this sort of technology generative ai, or just machine learning? Or it depends on the implementation?
I feel like most of the anger around ai is because gen ai has a bunch of harmful baggage, and I'm curious if this is an example of gen ai having a productive use case, or an example of ai being more useful outside of gen ai specifically
Based on their explanation, this is still using genAI. They talk about pre-processing the data and chunking it before it's sent to the inference model.
Based on what I read/saw and got out of it, I am real disappointed it looks like it's gonna be using genAI instead of another form of AI we've been using for transcription. Otherwise, sounds like trying to be a modern genAI version of that speech to text software I'd see ads for on TV. Possibly good for accessibility, but I'll wait and see after it comes out.
At least they claim the whole thing to be done offline after model installation and it's allegedly sandboxed with the audio data being stored in a memory buffer that allegedly will be erased after the session. So I'll have to wait and see how this all plays out before making more judgments on it.
More AI stuff as usual, waiting to see demonstration how this will work
From Wikipedia, the free encyclopedia
Linux is a family of open source Unix-like operating systems based on the Linux kernel, an operating system kernel first released on September 17, 1991 by Linus Torvalds. Linux is typically packaged in a Linux distribution (or distro for short).
Distributions include the Linux kernel and supporting system software and libraries, many of which are provided by the GNU Project. Many Linux distributions use the word "Linux" in their name, but the Free Software Foundation uses the name GNU/Linux to emphasize the importance of GNU software, causing some controversy.
Community icon by Alpár-Etele Méder, licensed under CC BY 3.0