hmm, I am getting Cloudflare internal eror 500 on audiogames, audiopub, podbean.
Sensitive content
reshared this
>boot windows partition
>Lenovo Vantage autostarts
>discord autostarts
>microsoft teams autostarts
>defender scans entire fucking disk
>several command windows pop up and disappear
>"compatibility telemetry" starts blaring my cpu and disk
>windows auto-downloads driver updates
>screen flickers and scaling changes several times
>onedrive starts blaring cpu and disk
>Lenovo Vantage autostarts
>discord autostarts
>microsoft teams autostarts
>defender scans entire fucking disk
>several command windows pop up and disappear
>"compatibility telemetry" starts blaring my cpu and disk
>windows auto-downloads driver updates
>screen flickers and scaling changes several times
>onedrive starts blaring cpu and disk
reshared this
After a few intense days with GPT Codex, it’s finally time to officially introduce Sara. #SARA or Simple Accessible Radio Automation, is a program primarily designed for radio presenters who host live shows.
Some notable features that are currently missing from other broadcast solutions accessible to blind users include:
a loop function – especially useful for intros,
a flexible number of playlists and players,
and a news editor module with the ability to embed audio clips that can be played while reading the news.
The application is 100% vibe-coded, and it’s definitely not recommended to test it during a live broadcast. It requires several sound cards to operate, and for now, I’m not planning to add any virtual audio routing like the one found in Radioboss or SPL. The program is and will remain open source, and perhaps someone interested in contributing will join the project.
In terms of screen reader compatibility, the main focus is on NVDA.
Happy testing: gitrls.com/michaldziwisz/sara/
reshared this
At 13, I joined a Wolfenstein 3d modding forum. 22 years later, my posts are still there - and still searchable.
The discussion I had in Discord 2 years ago?
Not so much.
We chose this. We can choose better.
blog.discourse.org/2025/11/the…
The Death of Community Memory
I recently spent forty minutes searching through Slack trying to find a technical decision — made eight months ago — about an app that 60% of my creative life depends on. I knew the conversation had happened. I remembered participating in it.JA Westenberg (Discourse)
feld likes this.
reshared this
Perspective Studio: The Easy, Accessible Way to Use AI on Your Computer taylorarndt.substack.com/p/per…
Perspective Studio: The Easy, Accessible Way to Use AI on Your Computer
Most AI tools today run in the cloud and require an internet connection, subscriptions, or logins.Taylor Arndt (Taylor’s Substack)
reshared this
Hi, the model select box for a new chat is empty even though I have several models installed with Ollama, do you know what's the problem here?
I am wondering what accessible SIP clients do you use on Windows and android? I would like to experiment with odorik.cz as there are plenty options, like IVR setup, call forwarding, voicemail, call recording and similar stuff which we will probably need for our community project.
Peter Vágner reshared this.
This thread is worth reading if you are a Google docs user.
Short version: When you export a document from Google Docs, Google replaces all your hyperlinks with links that allows Google to monitor the interactions of everyone you share your document with.
This hidden link replacement can potentially be used to build a model of your professional relations, where people who interact more with your content are considered a stronger relation.
Think about the implications.
fosstodon.org/@Joe_0237/111145…
Joe :ferris: :nixos: (@Joe_0237@fosstodon.org)
Today I found out that google docs infects html exports with spyware, no scripts, but links in your document are replaced with invisible google tracking redirects.Fosstodon
This entry was edited (1 week ago)
reshared this
I have just found a nice document scanning app for android that can do automatic edge detection, cropping, multipage scanning, OCR, PDF export and more.
It's called #makeacopy and it's using #tesseract engine to perform the OCR directly on the device with no internet connectivity requirement at all.
The app has almost full #a11y support for screen reader users in the sense that all the controls are clearly labelled and it's easy to navigate.
I can't resist and I have asked the developer if it would be doable to add a screen reader compatible notifications making the automatic edge detection somehow accessible as well.
Now I'd appreciate comments from low vision screen reader users, mobility trainers, people assisting other blind people or others who might be able to tell if my idea is viable and how much you like it?
Here is link to the github issue I have started: github.com/egdels/makeacopy/is…
Thanks for looking into it.
reshared this
@ondrosik I think iPhone has got something like that built-in already as I've tried several mainstream apps and they all seem to be providing guidance, including the Whatsapp's feature of sending photos of documents.
@Paweł Masarczyk Yes, the developer has not dismissed the idea and he is considering implementing it eventually. We have discussed it with additional comments in that github issue. CC @ondrosik
New WhatsApp web app, you can kindly go suck my ... Like, wow. I just recorded a voice message but pressing the send button does nothing. Yup, nothing. And generally, this is just garbage. Why? Why why why why. I'm going back to fucking Telegram and Elten Messages with this type of shit.
btw the fix to actually trigger the send button is going into focus mode and pressing it. And make sure you don't hit the view once radio button by accidant.
btw the fix to actually trigger the send button is going into focus mode and pressing it. And make sure you don't hit the view once radio button by accidant.
This entry was edited (1 week ago)
ondrosik reshared this.
WTF, and now they also fucked down the Audio quallity to what ever the heck of garbage this is now. Seriously?
Here's the most efficient way I've found to make a video from an image plus audio file that preserves quality.
- Get ffmpeg: Open the run box with windows+r, type "winget install ffmpeg" and press enter. ffmpeg will install itself.
- Open the folder on your PC where your image and audio files are stored. Press alt+d to go to the address bar.
- copy the below and paste it, Replacing the text in quotes with the names of your files. After pasting, press enter to run the command:
ffmpeg -threads 3 -hwaccel auto -r 1 -loop 1 -i "image.file" -i "audio.file" -c:v libx264 -preset ultrafast -x264opts opencl -vf scale=1280:720 -c:v libx264 -tune stillimage -c:a copy -shortest "video.mp4"
- A command prompt terminal should open and the encode should start. You can check the status by reading the bottom of the window.
- The window will close once the encode is complete. This could take anywhere from a few seconds to five-ten minutes. You should see a video file in the folder that is only a few MB larger than the audio+image files you started with.
This entry was edited (1 week ago)
reshared this
Make this a bat file on your path, start with image first, then audiofile, and you'll get audiofile.mp4 as an output.
@ffmpeg -threads 3 -hwaccel auto -r 1 -loop 1 -i %1 -i %2 -c:v libx264 -preset ultrafast -x264opts opencl -vf scale=1280:720 -c:v libx264 -tune stillimage -c:a copy -shortest "%~dpn2.mp4"
Andre Louis reshared this.
Zoom released New firmware for H1E. I am sending this again because i broke the first post.
- You can set imput gain during recording but i don't know exact buttons
- You can record and export to mp3
- You can record with AI noise isolator. It analyzes the noise during first 3 seconds to get noise profile. you can enable it with shortcut, the third button in the top row (button which marks during recording) or in rec settings. Not bad but it filters just high frequencies.
Do you guys testet this? @Andre Louis @Borris @Tomecki
zoomcorp.com/en/jp/handheld-re…
I have two complaints:
1) It seems that AI learning is just some kind of placebo. I tested it with various enviroments, noises, music... and I am getting the same result.
2) When you dont have headphones, you can set the imput gain volume too low so you actually record silence.
1) It seems that AI learning is just some kind of placebo. I tested it with various enviroments, noises, music... and I am getting the same result.
2) When you dont have headphones, you can set the imput gain volume too low so you actually record silence.
@tomecki@101010.pl @FreakyFwoof As to your second point, that's the way all their other recorders have worked for a while. Just bringing feature parity in line, I guess.
✨ Your new public_html folder is here, upload html and ~your-site is ready
~ public.monster 🐙
I'm pretty sure I'll regret making this, but upload your public monstrosity and let's see how it goes. Some fun may be had. I'll watch carefully for now 🕵️
This entry was edited (1 week ago)
reshared this
It is not like me to complain out here in social media land. But I feel it's time that I relay my experience with Envision, their customer service and the Solos glasses.
I received my glasses last Thursday. I was extremely excited as the product seemed like something that would work well for me. I watched all the videos, attended the webinars and read the user manual in advance.
I paired my glasses after charging them but after waiting for 3 hours, the glasses would not recognize the camera.
OK. I sent a message to customer service. I got the usual canned response that they would get back to me.
I got another copy of the user manual from customer service. I decided to ask how to return the glasses because there are no return instructions on the site or in the user manual. I did what was asked and contacted customer service to get the instructions.
The usual canned response ensued with another copy of the user manual.
After another much more firm email I was given a meeting with Karthik Cannan, who is one of the developers. After showing him that the glasses were, in fact, not working, he stated he needed to contact Solos in Hong Kong and would contact me within a few hours.
Well, many hours have passed. This meeting was on Monday and I've heard nothing since. I again emailed customer service, as well as a direct email to Mr. Cannan got the canned response but nothing more.,
This was not a small purchase. If this is the way a company treats people who have purchased products from them then I can't recommend anything they do, even if the product eventually works. All I want is to return the glasses, get my refund and be done with it. No harm, no foul. I hope to get this resolved soon or I'll be forced to report Envision to the Better Business bureau and take legal action. #LetsEnvision
I received my glasses last Thursday. I was extremely excited as the product seemed like something that would work well for me. I watched all the videos, attended the webinars and read the user manual in advance.
I paired my glasses after charging them but after waiting for 3 hours, the glasses would not recognize the camera.
OK. I sent a message to customer service. I got the usual canned response that they would get back to me.
I got another copy of the user manual from customer service. I decided to ask how to return the glasses because there are no return instructions on the site or in the user manual. I did what was asked and contacted customer service to get the instructions.
The usual canned response ensued with another copy of the user manual.
After another much more firm email I was given a meeting with Karthik Cannan, who is one of the developers. After showing him that the glasses were, in fact, not working, he stated he needed to contact Solos in Hong Kong and would contact me within a few hours.
Well, many hours have passed. This meeting was on Monday and I've heard nothing since. I again emailed customer service, as well as a direct email to Mr. Cannan got the canned response but nothing more.,
This was not a small purchase. If this is the way a company treats people who have purchased products from them then I can't recommend anything they do, even if the product eventually works. All I want is to return the glasses, get my refund and be done with it. No harm, no foul. I hope to get this resolved soon or I'll be forced to report Envision to the Better Business bureau and take legal action. #LetsEnvision
reshared this
Something is not good in this special-glasses world, I had a bad experience with Seleste before. As you are on CaneAndAble, you know who Jonathan Mosen is. If your problem is not solved, I strongly recommend you to write to the AccessOn podcast Jonathan is doing for NFB.
I just read that FFMPEG supports lufs normalization. That with combination with luveler and MrsWatson allows for nice automatic normalization of recordings. I can use Reaper and tweak everithing, but there are people who just need to get their recordings to relative consistent levels.Luveler teragonaudio/MrsWatson: A command-line VST plugin host
Peter Vágner reshared this.
So on a fresh Windows 11 install, I would need to now run at least 4 GitHub scripts just to make it behave and tamed down to the process plus ram usage of Windows 10. Unbelievable. I'm going to provide links here, you use these at your own risk. I'm also favoriting this post for myself so this is a bit for me, too.
All of these require that Defender's "realtime protection" and possibly "tampor protection" stay off. Defender is very yelly.
First thing I do: github.com/ShadowWhisperer/Rem…
This removes Microsoft Edge, and I use the WebView remover. If you use JAWS, this one might not be for you, since I believe JAWS relies on the webView.
2. Spicy: Remove defender. Not for everyone, but if you have a 3rd-party scanner and AV you use that's light, or just want to manually scan files and do daily quick scans with a tool of your choice, it's good. Using WSL will reinstall the hypervisor bits. It does a lot, so be cautious.
github.com/ionuttbara/windows-…
3. New to the list: github.com/zoicware/RemoveWind… - for getting rid of all AI features. Again, large script, read and consult carefully.
4. My own quick debloat Gist for LTSC: gist.github.com/tgeczy/2d847e2…
This does a lot more plumming removal, and disables your search box, so don't be surprised.
5. For search box: Open Menu: github.com/Open-Shell/Open-She… - works really well.
Bonus: github.com/Raphire/Win11Debloa…
Powershell script to debloat Windows 11 LTSC
Powershell script to debloat Windows 11 LTSC. GitHub Gist: instantly share code, notes, and snippets.Gist
ondrosik likes this.
reshared this
Thanks bunches! Taming is under way here - LOL! Glad I went with 32GB of RAM this time around. Guess next time it'll be 64GB to keep up with Microsoft, which is MacroHard on RAM.
@amir oh yeah, 32 GB is very respectible even now, I have 2 laptops here on Win11 with that amount. I think under 32, like 16, can also work because the OS scales back RAM usage a bit, by about 500-600 MB or more, so less aggressive caching policies on the 8 or 16 GB systems help mitigate anything you'd feel. Windows works in misterious ways. Ah wait...
Not to forget Winearo Tweaker and this one: amelabs.net/
Ameliorated
Completely transform your computer in minutes. Simply download a verified Playbook, or use your own, and run it in AME Beta.amelabs.net
I get an idea about this nice mashup. I started doing it and realized, is it possible that someone did it already? Yes, its already done. Now I have no idea how to procrastinate so its time to get back to real work. 
"Enola Gay Ti Amo" - OMD vs Ricchi E Poveri (mashup visualiser)
Enola Gay (Orchestral Manoeuvres in the Dark) vs Sarà Perché Ti Amo (Ricchi E Poveri)YouTube
My new favorite LLM trick.
Here's a link. Here's the JSON structure I need. Write me a Python script that takes this link and generates what I need.
15 minutes later, I have a working scraper.
This needs a sandboxed coding agent with network access approval (Openai Codex is perfect for this).
ondrosik reshared this.
Yes — the app is called Regionalny System Ostrzegania. The proxy option does sound promising. Are there any iOS apps that can capture that traffic, or will I have to set something up on the local network?
@midzi iOS apps won't have enough permissions to do this. You have to set up mitmproxy on your computer, you'll also need to install its TLS certificate (as a provisioning profile on the iPhone) so that it can intercept encrypted https traffic.
This assumes the app doesn't have certificate pinning. If it does, that's a lot more reverse engineering, an Android device would come in handy here.
Recently, I discovered Mynoise.net by @Stéphane (Dr. P). Sometimes, when I cant sleep or need background noise for boring work, I use various background soundscapes. Here is my article about Mynoise from the blind users experience. Long post follows:
Getting Started
- We won’t install or configure anything.
- Simply open the page mynoise.netmynoise.net in your web browser.
- Activate, for example, the link windSeaRainNoiseGenerator (wind, sea, and rain noise generator).
- Since browsers block autoplay by default, enable focus mode and press the letter p twice. The first press pauses playback, and the second resumes it. At this point, you should already hear the sound of the sea and wind noise.
Here you can also try other soundscapes, such as Rain on a Tent or Fireworks. I recommend opening the Full List. The soundscapes are organized under headings.
Playback Control
At this point, the possibilities of Mynoise are just beginning. Each player can be customized. Although the ambiance differs, the controls are always the same. Let’s take another look at the previously mentioned Wind, Sea, and Rain Noise.
Presets
Under the heading Presets, you’ll find buttons with various predefined settings. Activating them changes the sound’s characteristics. For instance, you can choose Breaking Waves or Irish Summer.
Under User Stories, you’ll find user comments. By activating a comment, the sound’s parameters adjust to the same configuration used by that commenter.
How to Create Your Own Background
Let’s say you want to manually adjust the balance between wind, waves, rain, and so on. Each player consists of ten sliders whose volume you can control. Here’s how:
- Enable focus mode in your screenreader.
- Press Enter to deselect all sliders.
- Choose one of the sliders using numbers 0–9.
- Adjust the volume with the + or – keys.
- Press Enter twice to deselect all sliders.
- Choose another slider (0–9).
- Adjust the volume again with + or –.
- If you’re satisfied but find the overall sound too loud, you can lower the volume with J and raise it with K.
- You can also generate a random setting by pressing the ? (question mark)key.
I Created a Beautiful Ambience—How Can I Return to It?
The author thought of that, too. Press B to find the Save as URL button. After pressing it, a URL containing your custom parameters will appear in an edit field. You can copy this link to your clipboard and save it as a bookmark. That’s how it works in Firefox. Chrome, on the other hand, will automatically reload the page with the new URL. You can share this URL as usual—for this article, I created This Noise as an example.
Can I Combine “Rain on a Tent” and a “Crackling Fireplace”?
Yes, and there are two ways to do it. If you’ve been experimenting with the site for a while, you might have opened one soundscape in one browser tab and another in a second tab. However, this setup is difficult to save or share. Fortunately, you can create a single page that combines multiple generators—up to ten soundscapes in one! This way, you can really make something like a Campfire in the Rain. Here’s how:
- Open your first soundscape, e.g. Rain on a Tent.
- Reset the sliders to zero and adjust them as you wish, as described earlier.
- Press C to save your current setting.
– In another tab, open the main Mynoise page and find your second soundscape.
– Open a new generator, for example Fireplace.
– Adjust everything again as you like.
– Press C again to save it.
– Use Shift+C to save all settings together.
– The page will ask for a name.
– Enter a name and confirm.
– You can now save the page as a bookmark or copy and share the URL.
Mobile App
Unfortunately, Mobile app is not accessible for blind users yet. You can still use Mynoise in your smartphone’s web browser. The easiest way is to share the custom links you prepared on your computer. That way, you’ll always have your favorite sounds at hand—for example, to help you sleep.
The author also offers online radios featuring some of the sounds. You can find them in the RadioBrowser database by searching for Mynoise. Additionally, there’s a podcast called Pomodoro Sessions, so you can enjoy your favorite soundscapes right in your podcast app.
Conclusion
The author is open to discussing accessibility. Thanks to this, button labels have already been added to the player pages. To celebrate finishing this article—and your reading it all the way through—you can listen to Fireworks.
Pomodoro Sessions
Podcast · myNoise · No ads, no talking. This podcast provides a wide range of background noises and music formatted for the Pomodoro technique: stop procrastinating and get the job done, 25 minutes at a time! These sounds won’t interfere with your fo…Spotify
Peter Vágner likes this.
Peter Vágner reshared this.
WhatsApp changes its terms to bar general-purpose chatbots from its platform. Meta said that the new chatbot use cases placed a lot of burden on its system with increased message volume and required a different kind of support, which the company wasn’t ready for. Really? I think that Meta just wants to force users to use Meta AI. techcrunch.com/2025/10/18/what…
WhatsApp changes its terms to bar general-purpose chatbots from its platform | TechCrunch
Meta's rationale behind this move is that WhatsApp Business API is designed for businesses serving customers rather than acting as a platform for chatbot distribution.Ivan Mehta (TechCrunch)
Having just purchased NKS presets for the Korg Collection 6 instruments this morning, I thought I'd sit down and go through some of them. I had to stop on this particular preset called 'German Piano Shimmer' and ended up playing a free-style version of the old 90's Dance anthem, Robert Miles 'Children' in my own particular way.
reshared this
Still very nice acoustic arrangement, that B section? I wasn't expecting! I apologise! I just could hear a drum line in my head and thought it was sound nice. The ending also is so pretty.
@TomGrant91 Would detract from it. Acoustic only this time. You can add drums if you feel so inclined but not for me. It's more fun showcasing the piano I feel.
HMM, Google notebook lm can import midi files. But that is all, no analysis, no replyes, nothing interesting.
Hmm, new wakeup alarm for me :D
Shuffle The Deck
I’m currently working on an interesting project. Last year, I met a former homeless man named Peter. He lost all his money to gambling. On the streets, he sold a street magazine, and later he started writing his own book. Today, most of his income comes from selling that book.
He wanted to create an audiobook. Since the budget was small, we did it like this: I gave him a Zoom H1N recorder. He locked himself in a relatively quiet room and gradually recorded the entire book. He sent me the raw material, which I ran through @Auphonic to remove background noise and room echo and to balance the loudness levels.
Now I just need to remove the mistakes and create the music background. It won’t be full studio quality — but honestly, I’ve heard “studio” recordings that sounded much worse than what we’re working on now.
He wanted to create an audiobook. Since the budget was small, we did it like this: I gave him a Zoom H1N recorder. He locked himself in a relatively quiet room and gradually recorded the entire book. He sent me the raw material, which I ran through @Auphonic to remove background noise and room echo and to balance the loudness levels.
Now I just need to remove the mistakes and create the music background. It won’t be full studio quality — but honestly, I’ve heard “studio” recordings that sounded much worse than what we’re working on now.
Peter Vágner likes this.
Peter Vágner reshared this.
Peter Vágner likes this.
Peter Vágner reshared this.
Stop Google from limiting APK file usage
ondrosik reshared this.
Keep Android open
keepandroidopen.org/
keepandroidopen.org/
Keep Android Open
Advocating for Android as a free, open platform for everyone to build apps on.Keep Android Open
reshared this
For those interested in airport and airplane communication, here are interesting recordings. Note to Screenreader users, it takes a while to load liveatc.net/recordings.php
LiveATC Recordings | LiveATC.net
LiveATC.Net Recordings - interesting air traffic communications captured by LiveATC userswww.liveatc.net
Peter Vágner reshared this.
My wife uses the Google speaker to find her phone. For some time now Google has refused to recognize my voice for that feature. But it seems I've found a way to trick it. audiopub.site/listen/e9777339-…
Peter Vágner likes this.
The AirPods Pro 3 flight problem
Link: basicappleguy.com/basicapplebl…
Discussion: news.ycombinator.com/item?id=4…
The AirPods Pro 3 Flight Problem
Testing Apple’s latest noise-cancelling earbuds at 39,000 feet reveals a potential flaw most users won’t notice, until they fly.BasicAppleGuy (Basic Apple Guy)
reshared this
Today I finally experienced Samsung’s HIYA spam blocking in action. It works perfectly — except that my food delivery courier couldn’t reach me. Clever guy though, he called again from a hidden number. I later found his repeated calls under “Other calls.” No idea why he ended up in the spam filter; it doesn’t seem like there’s a way to unblock him. Maybe adding him to my contacts would help…
This project looks promising. It looks like amazing replacement for Station playlist. Supports playlists, automation, looping. I sometimes play music during family events using SPL, maybe on next I will try this. Seems that my next procrastination moment will be with this. github.com/michaldziwisz/sara. Latest release at github.com/michaldziwisz/sara/…
GitHub - michaldziwisz/sara: Simple Accessible Radio Automation
Simple Accessible Radio Automation. Contribute to michaldziwisz/sara development by creating an account on GitHub.GitHub
reshared this
@ondrosik Oh, this looks fantastic and might even run on linux with some tweaking. Do I understand correctly for streaming to the server such as icecast I need something else for example edcast on windows?
yes, unfortunately, it doesn't support streaming yet. Also it seems that it doesn't support microphone imput.
Peter Vágner likes this.
@Piciok @pvagner I’m glad the project caught your interest. To be honest, I’ve put it on hold for now because I ran into problems I couldn’t quite overcome — at least not easily with AI. However, I’ll definitely come back to it someday, since I’ve always wanted to build my own live broadcasting system; I just lacked the coding skills. Yes, SARA will be designed primarily around the traditional model of a live radio DJ — using a mixer, sound card (preferably several), and absolutely no virtual mixers or anything like that.
Peter Vágner likes this.
@Piciok @pvagner I’m pleased to share that I’ve returned to working on SARA. The latest version is now hitting GitHub, featuring an improved audio engine, an NVDA add-on that tames the excessive verbosity of WX controls, and the ability to paste items from the clipboard — including folders. A word of caution, though: inserting large numbers of files is still quite slow for now.
In the next release, I’m planning something no other accessible broadcasting software offers: a dedicated playlist for reading news and easily triggering service sound cues — a feature typically found only in professional systems used for preparing news bulletins. gitrls.com/michaldziwisz/sara/ If there are any features you’re missing in programs like StationPlaylist or RadioBoss, be sure to let me know. Just keep in mind that for now SARA is meant to be operated mainly by a human — full automation will probably come later. :)
In the next release, I’m planning something no other accessible broadcasting software offers: a dedicated playlist for reading news and easily triggering service sound cues — a feature typically found only in professional systems used for preparing news bulletins. gitrls.com/michaldziwisz/sara/ If there are any features you’re missing in programs like StationPlaylist or RadioBoss, be sure to let me know. Just keep in mind that for now SARA is meant to be operated mainly by a human — full automation will probably come later. :)
Just curious: Can Station playlist use Asio imputs and outputs? Thei state that Asio is supported, but on my windows 11 with asio4all enabled SPL doesnt see asio devices.
Peter Vágner reshared this.
Yesterday I arrived from Cyprus back to Slovakia. I was there with my wife on a holyday. I feel littlebit tired because of 2-hour time difference. Now I am browsing all recordings and here is one from the airport: audiopub.site/listen/47ffd775-…
Enable This Windows Feature (Before It's Too Late) youtube.com/watch?v=KIgEww_rfS…Enable This New Windows Recovery Feature Now!
Sponsored: Get 90 Days of Bitdefender Premium Security - Absolutely Free: https://bitdefend.me/90ThioJoeOne of the worst feelings ever is when Windows fails ...YouTube
I really want to test this
For a few years I've been aware of this website that purports to be able to unlock shopping cart wheels using the speaker on your phone, but i finally had an excuse to try and and i remembered in the moment.A woman was outside the grocery store struggling to move her shopping car that was stuck because two of the wheels were locked
I remembered the website! So I put my phone near the wheels and played the sound. The wheels unlocked like magic. She was very happy. So cool.
One UI 8.0 Bluetooth Call Bug Silencing Screen Readers: What It Is and Current Workarounds accessibleandroid.com/one-ui-8…
One UI 8.0 Bluetooth Call Bug Silencing Screen Readers: What It Is and Current Workarounds - Accessible
When using a Bluetooth device such as a speaker or earbuds and raising the volume during a call, speech output eventually stops completely until the volume isKareen Kiwan (Accessible Android)
reshared this
I need to reinstall Windows on my Surface tablet, and I wanted to see if I could use AI to get into the boot menu and change the boot order to try USB first. I have a Chat GPT subscription, so I used voice mode with the video camera on, gave it some information about what I was doing, and let it be my eyes for a while.
I was delighted when the assistant actually knew how to get into the boot menu. It told me I could navigate the menu with my volume keys and select an option with the power key, much like navigating an Android phone’s recovery menu. I had to remind it I was totally blind a couple of times, but eventually, it helped me select USB as the first boot device and choose the “Exit and Restart” option. This was quite a few steps, and I was genuinely impressed at how easily I was able to fix this problem without interacting with a real person.
I eagerly waited for the Windows installer to come up. It never did.
So I used OCR on the screen, and discovered that not a single thing chat GPT claimed had happened … had actually happened.
It didn’t hallucinate just one thing. It hallucinated an entire multi-step interaction with the firmware of my tablet. It basically experienced a break from reality for about two minutes and started describing what it thought should happen, with no regard for what was actually happening.
Last week, the same app helped me learn the control panel of my heated mattress pad. It does work sometimes. But today, it led me on the wildest goose chase I’ve ever been on. I was actually trying to boot from the SD card, and as it turns out, that’s not even an option in the boot menu. But I made the mistake of teling Chat GPT exactly what I was trying to do, so it had all the material it needed to hallucinate a complex interaction convincingly.
Never let yourself forget the all important “A” in “AI”. That intelligence is not artificial as in “synthetic”, it’s artificial as in “pretend”. No LLM has the slightest idea of what it’s doing or saying. The companies that create these models have the all-important task of trying to make their intelligence more convincing than every other company.
That means they work most of the time, but the rest of the time, they will confidently lie. And that lie might be a missing digit, or it might be a whole entire interaction with a device.
I called Aira and got it sorted. I actually had to touch the arrows on the touchscreen to rearrange the boot order. Yes, I had a keyboard connected. No, there is no documented way to rearrange boot devices on a surface using the keyboard. Yes, everythyng about this is moronic. But, it’s done, and once I get Windows on my USB device, I’m pretty sure it will boot because an actual human told me so.
I can’t even begin to enumerate the possible clusterfucks that could arise from AI weaving such complex webs of lies. Do not use this shit for anything mission-critical. Ever. Even if it told the truth the last 99 times. Eventually, it will lie. When it does, you’ll have no idea.
I was delighted when the assistant actually knew how to get into the boot menu. It told me I could navigate the menu with my volume keys and select an option with the power key, much like navigating an Android phone’s recovery menu. I had to remind it I was totally blind a couple of times, but eventually, it helped me select USB as the first boot device and choose the “Exit and Restart” option. This was quite a few steps, and I was genuinely impressed at how easily I was able to fix this problem without interacting with a real person.
I eagerly waited for the Windows installer to come up. It never did.
So I used OCR on the screen, and discovered that not a single thing chat GPT claimed had happened … had actually happened.
It didn’t hallucinate just one thing. It hallucinated an entire multi-step interaction with the firmware of my tablet. It basically experienced a break from reality for about two minutes and started describing what it thought should happen, with no regard for what was actually happening.
Last week, the same app helped me learn the control panel of my heated mattress pad. It does work sometimes. But today, it led me on the wildest goose chase I’ve ever been on. I was actually trying to boot from the SD card, and as it turns out, that’s not even an option in the boot menu. But I made the mistake of teling Chat GPT exactly what I was trying to do, so it had all the material it needed to hallucinate a complex interaction convincingly.
Never let yourself forget the all important “A” in “AI”. That intelligence is not artificial as in “synthetic”, it’s artificial as in “pretend”. No LLM has the slightest idea of what it’s doing or saying. The companies that create these models have the all-important task of trying to make their intelligence more convincing than every other company.
That means they work most of the time, but the rest of the time, they will confidently lie. And that lie might be a missing digit, or it might be a whole entire interaction with a device.
I called Aira and got it sorted. I actually had to touch the arrows on the touchscreen to rearrange the boot order. Yes, I had a keyboard connected. No, there is no documented way to rearrange boot devices on a surface using the keyboard. Yes, everythyng about this is moronic. But, it’s done, and once I get Windows on my USB device, I’m pretty sure it will boot because an actual human told me so.
I can’t even begin to enumerate the possible clusterfucks that could arise from AI weaving such complex webs of lies. Do not use this shit for anything mission-critical. Ever. Even if it told the truth the last 99 times. Eventually, it will lie. When it does, you’ll have no idea.
This entry was edited (1 month ago)
reshared this
Ahaha, totally reminds me of me trying to get secure boot disabled with GPT - it confidently kept saying I turned it off, that I'm in the advanced tab and just press arrow down a few times to find it. Of course, I was in the boot tab, and I later resorted to asking it more limited questions because the moment I would give away what I wanted to do, it would get way worse. No asking it, "Look for somethhing with the label secure boot and tell me how many arrow downs it is." Had a similar experience trying to get it to read the MAC address off a router, but at least there I could use something more local like Seeing AI and get an answer that I knew was as accurate as the ability I have to keep a good angle for it to read it. I think that's the other problem with GPT or other AI: It doesn't inform you like a human (always) that you need to move or tilt your phone. I've had it help me with that a few times, but unless you specifically ask it whether your phone needs adjusting, it would rather make up the thing than tell you.
Zach Bennoui reshared this.
I wonder if we could finetune a small VLLM specifically to describe photos and read text as accurately as possible without being able to do much else? I think the issue is these models are designed to be generalists, so they're not really good at any one particular thing.
This entry was edited (2 weeks ago)
I hate AI customer service support. You need to fix or solve something, call the customer service and AI agent tells you everithing what you already did. Finally you get to the real person but it takes longer time. To be clear, I use CHATGPT, Auphonic and other AI tools but for customer support it doesn't work.
Interesting, gps module on my android phone works also in airplane mode. For some reason I thought tha gps is disabled in airplane too. So I can check the speed during my next flight.
Peter Vágner likes this.
Jamie Teh
in reply to lesbian klonoa • • •