@menelion It’s an AI assistent, you just talk to it and ask it stuff and it answers back. You can set it up to use Gemini or GPT4O, or you can roll your own. right now I have it set to Gemini 1.5 flash which is Google’s newest low latency thing. But since it is Eleven Labs that means I can use the professional voice I already have with them. If I wanted to I could train it to be more specific, like, if I was a famous author I could give it my books and then people could ask it about them, that kind of thing.
@menelion It is a new feature so right now they are paying for the calls, but I suspect that will change as they work out the details.
I am not sure what language it is in. The thing that makes it so cool is you literally just hit “create,” it asks you what parameters you want to set for it like the voice you want, how createive it can get, the model you want, additional data sources, etc. and it just creates it for you. I haven’t written a line of code in my life and I got it working in five minutes.
@menelion Oh wow, if you have a twillio number you can link it and have a phone number people can use to call the agent! That’s so cool. Never going to use it, but that’s amazing. You could spin up a whole customer service agent with your own voice in like fifteen minutes. Crazy!
Hi Drew! It's me again, again a request about WhatsApp and this time a feature request. It would be extremely nice if there was a setting that would hide phone numbers and show/announce only names. Especially in larger groups, where there are lots of people I'll never add to contacts, it's so annoying to hear something like: "~John doe +353 123456789, replying to ~Jane Roe +44 999 1234567". thanks!
@menelion Hey thanks for the feedback! BTW, my rant about accessibility testers acting like ADA inspectors was completely unrelated to this, totally different thing.
The phone numbers are necessary in order to verify the person you are talking to, since anyone can set whatever name they want in their profile. Bad actors could use this to impersonate people you know.
I agree that it’s annoying though from the screenreader perspective, and we’ve gotten a fair bit of feedback to filter them out. We’ve been able to balance the concerns on mobile by including the number in the custom action hint when you swipe down from the bubble (E.G. “view contact info for +1 (202) 555-1111”) It’s more complicated for desktop though.
As a best practice, you are supposed to convey any information that is relevant to the screenreader for it/the user to decide how to handle it. I think this is an area where it makes more sense for a screenreader add-on or script perform this function so it is under the user’s control.
All that to say, it’s tricky, but I think there is a path forward here that will give the users what they want without security compromises.
Guys check out this new project, it's a cross platform app that lets you experiment and train open source LLMs without any code. It's based on Electron and open source, but I haven't yet looked at accessibility. github.com/transformerlab/tran…
Open Source Application for Advanced LLM Engineering: interact, train, fine-tune, and evaluate large language models on your own computer. - transformerlab/transformerlab-app
The irony: OpenAI—which faces multiple lawsuits for using content without permission—accuses China's DeepSeek of basically copying from ChatGPT to train its AI models. pcmag.com/news/openai-deepseek…
OpenAI—which faces multiple lawsuits for using content without permission—accuses DeepSeek of 'distillation,' a technique that use larger models' outputs to improve smaller ones.
“Nobody asked for NFTs or the fucking Metaverse. Nobody asked for lying chatbots instead of getting to reach an actual support person that could solve your problems.”
And:“…people are challenging the notion that we all have to do AI now. Because we don’t. It’s a choice. A choice that mostly benefits monopolists.”
I am quoted at length in a new article on Ars Technica (a website I have been reading for years and years) on an in-depth article on tarpits for AI crawlers.
Nvidia RTX 5080 Reviews Fail to Impress Gamers "That’s due to the RTX 5080 failing to surpass the RTX 4090 and only offering slight improvements over the RTX 4080 and RTX 4080 Super." (Still no talk of whether it runs less hot than the prior generation, so far none of these gaming sites covered that one) tipranks.com/news/nvidia-rtx-5…
Ignore the title, but look inside. I think it's an interesting point of view. Trump inauguration: The real reason Elon Musk, Mark Zuckerberg, and Jeff Bezos support Trump | Vox vox.com/future-perfect/395646/…
For billionaires inspired by science fiction, it’s about using crypto and network states to undermine democracy — not just lower taxes and friendly regulations.
Elena má rakovinu. Tvrdí výzva na sociálních sítích. Pokud jí přispějete na péči, jakou máte jistotu, že peníze rodina dostane? A je tenhle příběh vůbec skutečný? Pátrala, ptala se, a poví Jana Magdoňová, spoluautorka rozhlasového pořadu Antivirus.
#deepseekr1 Pro ty, jako jsem já, co tu online verzi nezkoušeli. Tohle je fantastické. Podobá se to mimochodem mé zkušenosti s lokální verzí LLM DeepSeek-R1: Tendence přemýšlet nahlas o tom, co "uživatel" (ano, ono to o vás myslí nahlas ve třetí osobě) chce:
"DeepSeek-R1 did not rush to a conclusion. Instead, it began to deliberate — openly, methodically, and exhaustively. It did not simply think of a number but dissected the task with a surgeon's precision and a philosopher's hesitation. Each clause of the instruction became a branch on the tree of reasoning, and every ambiguity was an opportunity to explore possibilities, reevaluate steps, and entertain contingencies."
"“Don’t overcomplicate it,” the machine echoed as if repeating these words might anchor it against the tides of overthinking. Yet, even as it sought simplicity, it spiraled into complexity. “A bit of a challenge,” it pondered. What, after all, constitutes a challenge? A number that is mathematically rare? One positioned oddly in the range of 1 to 100? Perhaps a prime number, or one nestled between decoys?"
I really like my Fantic MiniStreak fnatic.com/products/mini-strea… but now I want to use it with a Mac and in my dreams I want an almost 1:1 mapping to Mac keys but at the same time I'm too lazy to remap keys. Anyone who went the same path?
Unleash your speed and precision with the Mini Streak RGB Tenkeyless Speed Silver keyboard. Dominate the competition and enjoy free shipping over 100EUR/USD/GBP!
@gagliardi_vale hahaha ... don't worry, @Codeberg is a Git hosting provider that uses @forgejo (a self-hosted lightweight software forge) An exclusively Free Software alternative to GitHub and GitLab.
I'm really looking forward to seeing how the EchoVision Glasses from Agiga develops. I've pre-ordered and am hoping that they fulfill their promise... Read the alt text they shared with me in the alt text of this image since it wasn't in their newsletter. Don't forget to check out the interview I had with them, too! buff.ly/4gUnFNL
The EvhoVision smart glasses is a wearable that looks like a pair of regular glasses, but features multiple accessibility tools and AI made especially for th...
Postmedia is only 00.37% Canadian-owned. One-third of one percent. This explains the many columns in their pages right now promoting US annexation of Canada or Alberta succession to the US. #cdnpoli #postmedia #media
Looking forward to a busy Sunday at @fosdem with @matthew delivering ‘The Road to Mainstream Matrix’ at 12pm, followed by ‘Matrix State of the Union’ together with @Amandine at 1pm.
Attached: 2 images
The next Outreachy internship cohort application starts in a few days.
FOSS communities, coordinators, mentors, and applicants, are you ready?
Dear FOSS communities, coordinators, and mentors.
By default, Jieshuo includes a set of functions assigned to some gestures, while others remain unassigned. Although modifying the default gesture scheme is
![a sketch of a website for Company
there is a VERY BIG BANNER in the middle with text that says "use our shit, you should really use our shit, please use our shit" and a button that says "button text" and an image that says "[image of a guy being happy because he's a stock photo]"
the text in red says "i don't read any of this shit"
"i actually just go here" and it points to the top right log in and sign up buttons](https://fedi.ml/photo/preview/640/633078 "a sketch of a website for Company
there is a VERY BIG BANNER in the middle with text that says \"use our shit, you should really use our shit, please use our shit\" and a button that says \"button text\" and an image that says \"[image of a guy being happy because he's a stock photo]\"
the text in red says \"i don't read any of this shit\"
\"i actually just go here\" and it points to the top right log in and sign up buttons")
André Polykanine
in reply to Drew Mochak • • •Drew Mochak
in reply to André Polykanine • • •André Polykanine
in reply to Drew Mochak • • •Drew Mochak
in reply to André Polykanine • • •@menelion It is a new feature so right now they are paying for the calls, but I suspect that will change as they work out the details.
I am not sure what language it is in. The thing that makes it so cool is you literally just hit “create,” it asks you what parameters you want to set for it like the voice you want, how createive it can get, the model you want, additional data sources, etc. and it just creates it for you. I haven’t written a line of code in my life and I got it working in five minutes.
André Polykanine
in reply to Drew Mochak • • •Drew Mochak
in reply to André Polykanine • • •André Polykanine
in reply to Drew Mochak • • •victor tsaran
in reply to Drew Mochak • • •Zach Bennoui
in reply to Drew Mochak • • •André Polykanine
Unknown parent • • •André Polykanine
Unknown parent • • •Drew Mochak
in reply to André Polykanine • • •@menelion Hey thanks for the feedback! BTW, my rant about accessibility testers acting like ADA inspectors was completely unrelated to this, totally different thing.
The phone numbers are necessary in order to verify the person you are talking to, since anyone can set whatever name they want in their profile. Bad actors could use this to impersonate people you know.
I agree that it’s annoying though from the screenreader perspective, and we’ve gotten a fair bit of feedback to filter them out. We’ve been able to balance the concerns on mobile by including the number in the custom action hint when you swipe down from the bubble (E.G. “view contact info for +1 (202) 555-1111”) It’s more complicated for desktop though.
As a best practice, you are supposed to convey any information that is relevant to the screenreader for it/the user to decide how to handle it. I think this is an area where it makes more sense for a screenreader add-on or script perform this function so it is under the user’s control.
All that to say, it’s tricky, but I think there is a path forward here that will give the users what they want without security compromises.