Alibaba's Qwen group just shipped Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M - two Apache 2 licensed LLMs with an impressive 1 million token context limit!
Here are my notes, including my so-far unsuccessful attempts to run large context prompts on my Mac simonwillison.net/2025/Jan/26/…
Very significant new release from Alibaba's Qwen team. Their openly licensed (sometimes Apache 2, sometimes Qwen license, I've had trouble keeping up) Qwen 2.5 LLM previously had an input token …
This was a great find. If anyone knows when this was recorded, I'd love to know. This is "an Hour With Jon Pertwee" and it's just him, in front of a live audience, telling stories about his life. He mentions traveling to a Doctor Who convention so it's definitely sometime past 1970? #DoctorWho ia800608.us.archive.org/14/ite…
#NYTimes: "Colombia’s President Says He Will Not Accept U.S. Military Deportation Flights
Gustavo Petro said on X that the United States should not treat Colombian migrants as criminals and that he had already turned away U.S. military flights carrying deportees."
well, it didn't last long. Trump threatened them with 25% tariffs and the Colombian authorities announced they would send their own planes to the US to pick up Colombian nationals.
DJs.social is a friendly mastodon instance for DJs of any kind. Discuss your selections, techniques, equipment, and gigs. Share your mixes, videos and inspiration.
Meta's new Live AI is an early access feature that allows for continuous video chat with the AI without having to repeatedly say, "hey Meta". You can ask abo...
:Bother" as Mole used to say in Wind in the Willows, when my family asked me what I wanted for my Birthday last week, I didn't think of the Talking Air Frier sigh smile. Looks like another purchase is coming up soon grin.
⏰ What time is it? For Braille users, checking the time can be a breeze with JAWS! 🧑🦯 In this episode, we break down how to use JAWS with Braille displays to keep track of time effortlessly. 🎧 Tune in now: unmute.show #JAWS #Braille #Accessibility
Welcome to Unmute Presents Where Technology Meets Community At Unmute Presents, we connect blind and sight-impaired individuals through engaging podcasts and live discussions.
The disappearance of the White House Accessibility Statement from the White House website on President Trump’s first day in office is more than symbolic – it signals a larger problem.
I've been using Pop! OS on my laptop because of the touchscreen support so I've been with it for the last year, but I used Manjaro for at least 3 years and I can say that still my favorite. However I'm planning to try MX Linux, heard a lot of good things 'bout it.
It may be late, but combined with CRLite (and its other Web PKI progress), it may soon be the browser with the most robust Web PKI support.
While I would still say that Chromium generally wins on the security front, I’m happy to see the gap narrow with time and to see Firefox occasionally inch ahead in some areas.
Firefox 136 looks poised to enforce Certificate Transparency. It may be late, but combined with CRLite (and its other Web PKI progress), it seems to now be the
🔥 Amerika, země kde nechcete být ženou: Trump po projevu na Pochodu pro život podepsal plošnou protipotratovou politiku. Mezi postižené patří i zahraniční zdravotnické organizace, které distribuují antikoncepci a pomáhají v boji proti HIV.
"Aye, the sirens," says the lighthouse keeper, picking a bit of tobacco from his beard. "Some nights, when the wind is just so, I swear I can hear 'em singing." "What do they sing?" I prod him. A faraway look in his eyes, he says in a voice barely above a whisper, "Tubthumping."
Dear #Android #App #Developers, as it still happens far too often (no naming, no shaming! 💩 happens to everyone of us) a reminder to take good care of your #signing keys – and also take precautions for the case that your keystore might get lost. Please take a look at: f-droid.org/2023/09/03/reprodu… where I outline this topic.
Earlier this year, we reported about our progress concerning reproduciblebuilds. Meanwhile,more and more apps are using this; you can find some statisticsher...
@S1m Key rotation does no longer work at F-Droid.org, but it does at IzzyOnDroid (as we implemented the suggested patches instead of accepting their implementation of the "POC fix" back then). If Key rotation is used, no notifications are needed; IIRC, Android handles that (we have only 1 such app yet). And establishing RB here does not require it either, as we only ship the APKs signed by their resp. devs to begin with (RB runs on a "parallel track" here).
@S1m I have my doubts towards the former (that would mean rolling back their implementation, at least in parts, and using the suggested patches instead, which they rejected. The argument was that f-droid.org itself does not need key rotation, IIRC.). And as long as we still use fdroidserver, the only way we can notify is via the inlined per-release changelogs (aka "Fastlane changelogs"), which is what we do.
@S1m which needs to be implemented serverside (fdroidserver writing the index) AND clientside (to show it). Without the index itself supporting it, there's nothing the clients can do. So: gitlab.com/fdroid/fdroidserver… ? gitlab.com/fdroid/fdroidclient… ? Does not look like this will happen.
I'll try to open issues for more specific messages then: metadata to warn about removed application, and metadata to warn about application that need to be reinstalled
@S1m Remember that WE cannot implement those. Everything that needs changes to the index, is out of our hands – at least as long as we're still bound to fdroidserver. So such issues would need to be filed *there* – with the tools that generate the indexes. Our client devs would surely pick up those data once available.
@S1m the f-droid client doesn't handle key rotation. I think neo store and droid-ify allow disabling the key fingerprint check allowing android to handle the rotation but all clients will by default consider the key incompatible and not offer an update as the index format itself does not support key rotation even when generated using an implementation that doesn't flat out reject all APKs with rotated keys like fdroidserver does
@obfusk @S1m Confirmed that NeoStore can handle that. Just checked: Droid-ify had a new release this year already, so it should now handle that as well by now (last time I've checked it was implemented but not yet published, now it should be both). I cannot tell about other clients or how they might handle it.
But sure, let's randomly assign incredibly loaded clinical diagnoses onto individuals who may have many valid reasons for not wanting to engage in-person with people, particularly ones who make damning evidence-free judgements of others and share them in a newspaper
Locatiegegevens vergaard door ruim honderd Nederlandse apps staan online te koop. Het gaan en staan van hun miljoenen gebruikers is zonder hun medeweten te volgen. ‘Hartstikke illegaal’, zeggen deskundigen.
Did you know that you can add and use "ephemeral profiles" where all messages get deleted after a set time? The "auto-delete from device" will keep your profile clean. It's a one sided ephemerality that you decide about, not your chat partners. Good for coordinating things without digital traces and where only "the current day" or "week" is important but not all the past history. Saved messages stay, though, until you explicitely delete them. #ephemeral #deltachat
I want you to know that your CAPTCHA is inaccessible for disabled people: - blind people can't see the clock; - visually impaired people may not be able to see it correctly too; - dyscalculic people (like me!!) have difficulties reading a clock (moreover when all numbers are not written on it)!
Various approaches have been employed over many years to distinguish human users of web sites from robots. The traditional CAPTCHA approach asking users to identify obscured text in an image remains common, but other approaches have emerged.
We currently do not have an accessibility policy but we have brought this up with the team to consider. Regarding accessibility, our team is constantly working to improve it - including the captcha :)
Osvětim je za Ostravou. Na Krakov. Návštěva je nezapomenutelná. Měla jsem štěstí. Byla jsem částí školáků z Israele s vlajkou. Utěšovali mě. Ještě větší tábor je sousední Březinka. Průmyslová továrna na vraždění lidí. Jak někdo dnes může volit neonacisty je záhada?
As an aside, I also fondly remember your Bitstream jingle, as I was a subscriber to that magazine. I wonder what happened to Peter Ciali. Is he still with us?
Update: Thanks @pitermach showing a great demo that it's actually Mist World Upsampling to 48 in this demo, not NVDA downsampling to 16! I stitched together an audio file showing you how bad it is at ignoring the setting of -1 as the output. Instead #NVDASR tries to be too smart, enumerate the list and gather which you have set as your sound mapper output, and explicitly call that sound device when passing to the TTS outputs. I updated this to add a little more at the end and show how Mist World treats audio output switching properly, that I now know is not proper. Good night, Mastodon. This really ruined my weekend at first, until that amazing demo in my mentions by @pitermach clarified things. :) Update: People are asking, "how can I tell?" Listen for the sharpness of S's and other consonants. If you have the ear you'll notice.
What you're hearing isn't actually downsampling to 16, it's aliasing artifacts introduced by whatever resampling algorithm Mistworld's audio library is using. Vocalizer actually runs at a native 22 KHZ as far as I know. I recorded a quick demo of what it sounds like when you bring a 22 KHZ file to 48 KHZ with a low quality resampling algorithm versus a file that's actually at 16 KHZ.
@JamminJerry There isn't much of a difference between the old default 64 point sync interpolation and r8brain, but a more pronounced difference with point sampling and linear interpolation there's a bit more high end. Something else I just remembered is back in the day Klango also used to do this, so any voice you used with it would get resampled up like this with noticeable aliasing.
super informative, wow! I wonder if the default output set within Windows sound pannel to 48K causes the upsampling to 48 and not keeping at 44100, how odd. It could be that with the sound mapper Windows always forces its own sampling rate rather than sticking with the one set as playback in the program. But I'm not 100% sure. (Nope, not this, did a test run by force-changing a Bluetooth A2DP driver to 44) It's definitely odd that when you choose an audio device directly it's correctly sampling in the game though, and maybe Klango did the same thing.
Unless you're using ASIO and/or exclusive mode, some degree of resampling is unavoidable. Windows opens the device with a particular configuration of sample rate, number of channels, etc. and audio from applications is adapted as needed to match so you get a mix. If three applications are simultaneously sending audio output to the same device, the device is only opened once and receives the sum total of those sources.
It's also odd that there's a difference because with WASAPI, which even legacy WinMM now uses behind the scenes, there isn't really a separate sound mapper device. You ask the system for the default endpoint, it tells you which that is (the real endpoint, not some sound mapper thing) and then you open that endpoint directly. Thus, there should be no difference in how the device is actually opened. I guess it's possible the app makes different decisions about resampling based on whether the user wants to use the default device or not, but why would it do that?
I'll also flag there are different resampling flags you can specify when opening the device with WASAPI. By default, Windows uses a really crappy resampling algorithm, but you can request a better one. I learned this the hard way when I was first implementing WASAPI in NVDA. But again, why would the app fail to specify the correct flag based on which device the user chose?
and... if you want this aliasing, use Zdsr. I am trying to argue with people that this aliasing is bad. No, they know better, because they have more experience, and stuff like that... If i see that it cannot be hear for a long time... i stop using it.
yeah, I do think that the new R8Brain algo mentioned there does a lot better job at still making the voice have that higher crisp quality but not so much the sharpness on the actual consonants, which to me feels like the best of both worlds. You get a lot higher quality to the ear but also don't get those weird artifacts the older ways of upsampling to 44 or 48 introduces. I don't think people who don't like it are wrong, especially for some minds, sharper noises like that in the audio can really stand out and become annoying or a headache.
Because of this harsh aliasing, i start to loss concentration while listening to speech. I don't know, but RHVoice uses its original sample rates IN NVDA or using sapi. also a screen reader which is not doing good with sample rates is jfw.
@asael I wonder if we'll ever get a true TTS that's not 22050 but true 44.1K sampling rate. Now I think I'm on the hunt for that. My guess is the newer AI voices might be the first of their kind this way if so. It's interesting because 22050 in actuality is 11025. The Nyquist frequency (or Nyquist limit) refers to the highest frequency that can be accurately represented by a given sample rate. It is half of the sample rate. The reason for this is due to how digital sampling works: you need at least two samples per cycle of the waveform to fully capture its shape. This is known as the Nyquist-Shannon sampling theorem. So really, any TTS claiming to be 22050 HZ is really just 11025, and any TTS claiming to be 11025 is just 5.5 K-hertz, youch
@asael We had one, for a very brief moment, Innoetics. But they got bought out by Samsung. I think there's also a bulgarian TTS that runs at 44, saw someone use it once but not sure what it's called.
I have (had?) the Innoetics John voice. It's probably still installed on my old laptop. It was kind of odd and weirdly concatenated, but definitely had highs that nothing else did.
I've thought of making a voice of myself in RHVoice but I'm unsure about the intonation, I often talk iwth a higher inflection and want that to be reflected, and haven't quite figured out the ridiculously complicated pipeline. Plus I've only got WSL to train it so it will be sloooooow! @valiant8086 @BorrisInABox @pitermach @Tamasg Oh, has that replaced the one they'll ship for the NVDA add-on? I only see one version of SLT on rhvoice.org
@BorrisInABox @asael OMG. This was a thing? What year did that all get created? Feel like I've missed like, a major milestone in speech history. LOL. It may be that using actual 44K data is too large in size, so that's why even the human voices by companies such as Nuance stuck with 22050, as it's tolerable enough to not be an AM radio but not high quality to be even like an FM signal could be.
@asael @BorrisInABox Believe it or not I still have it here, yay for a Windows install on its 9th year now. Fun fact, that John voice was created from the voice of Jon St. John, aka the guy who voiced Duke Nukem, sadly not doing the Duke voice in this case. But that single reason is why we used it for a very long time either for reading chats or games while streaming with @talon. Here's a quick demo. Not a super responsive voice, but yeah there's a lot of highs.
@asael @BorrisInABox @talon wow yeah, that dfor sure has some lag, although have heard worse. Sounds more like what you'd get on a Piper voice. But yeah, the quality in this is for sure what I like about the up-sampled Vocalizer, and there's none of that sharpness crap at the ends of certain letters people can notice, wow.
@BorrisInABox @talon btw, there is something new about upcoming NVDA. The chinese dev, who is developing natural sapi adapter, started making improvements to speech protocols. He is going to add the silence trimming for all tts, making them responsive.
@asael @BorrisInABox @talon Oh neat. That's definitely something that's giving ZDSR a decent performance edge over NVDA with some SAPI voices it makes a very noticeable improvement.
@masonasons @asael wow did they ever do any demos with actual Innoetics voices to show that? Never heard of that company, but it would make sense why or how Samsung got such high quality voices and a major improvement from the sing-song Samsung TTS they had to the way better voices used now.
Comedian Bill Bailey reimagines the Doctor Who theme as Jacques Brel-esque Belgian jazz. Taken from his DVD 'Bill Bailey's Remarkable Guide to the Orchestra'...
Matt Campbell
in reply to Simon Willison • • •Simon Willison
in reply to Matt Campbell • • •