Nice, been working on this post and project for a bit over a month and a half, with the big move in between of course. Glad it finally went live.
I built an open source image description service which will eventually get its own NVDA addon, and is of course hosted on fly.io. fly.io/blog/llm-image-descript…
I, for the first time in 11 years, have a new physical address.
There are a lot of stories I could tell about that and the process of getting it, but holy shit I got out of practice. I used to switch addresses every couple years. I guess I forgot how awkward and anxiety-inducing the first few nights alone at a new place are. I don't know any of this background noise yet.
To be clear, I haven't heard anything alarming. It's just, places sound very different even if they're quiet, and I have 11 years of experience telling me dysfunctional home sounds like something else. Now I have to learn what maybe-hopefully-functional home sounds like.
So far so good with the new place. We'll see how I feel in a few months, but as long as we don't deteriorate to management actively gaslighting and deflecting me, it'll be a massive improvement. 😀
Oh and then there's the "I'll put my phone where I normally do, on my desk. But hey, there's no desk. Let me drop it somewhere that makes sense in the moment, do 15 things and hey, where'd I leave my hone?" Replace "phone" with half a dozen other things, and that's also my life of late. I'll figure all this shit out eventually I'm sure but right now the cognitive load is crushing.
@objectinspace That's what speech-to-text is--converting speech to text for the model to process. Neat, wasn't sure if it was mostly a spoken interface or something else.
And here's me, stupid, thinking this move would be a lot easier because I had more income. It is, but holy hell have things gone to shit since I did this last in 2013.
Like the apartment I'm withdrawing my application from? They don't know if they can or will be able to get me an accessible copy of my 37-page lease. In 2024.
Deleted a much longer, rantier post about the legal shenanigans my partner noticed while she read parts of this inaccessible lease to me. I'll just say they switched a guarantor application to a resident application without my consent to work around a credit score requirement. Yeah, bullet fucking dodged. I did not consent to that individual as a resident, only as a guarantor, and the application we filled out made that very explicit.
If only I hadn't put just about everything I own into storage already. A dumpster fire at the shit show, that's my life right now. Hope yours is better.
Does Friendica not support uploading audio? Tried uploading an MP3 but both times I couldn't find the uploaded file in the posting interface.
I'm also using Windows and noticed I had to select all file types to find my MP3 to upload. So maybe Friendica doesn't disallow non-images but the file selection makes it seem like it might.
I've never messed with GPUs under Linux. Anyone have tips on how to stress-test one, or otherwise put it through its paces, that aren't gaming or mining crypto?
I realize the latter might be OK if I did it only in the context of a stress test and for a limited time, but aside from wanting to stay away from crypto on principle, this is for work and we have systems actively scanning for and flagging it.
Context is that I'm stress-testing our new GPU offering at work to find bugs. It doesn't necessarily have to be a stress-test, but I'm trying to actively break things in a way that won't trigger our scans but I don't know the first thing about this stuff.
And FWIW we've already generally run LLMs and such, so the lowest hanging fruit is already picked.
Hey blind Android users, anyone have a favorite more advanced GPS navigation app?
In particular, I'm looking for something that will let me save my own points and navigate back to them, with both directions and bearing/distance for cases where Google Maps tries to do a dumb and take me along small access roads and such when really I just need to hang a left and walk for about 100 feet, but the app doesn't give me that info because it's too busy holding my hand.
Likely moving to a new area next week, there's a lot of cool stuff to walk to, and I'd like to just mark my building entrance so I can return after exploring things. But so many of the apps I'm finding want to hold my hand more than is necessary, and that sucks.
Blind folks (or I guess anyone really): how do you take screenshots accessibly and safely under Windows 11?
By "accessibly" I mean being able to just get image content of the current window easily. There's some sort of snipping tool but based on the name I feel like that's probably based on visually highlighting what you want to screenshot, which is a non-starter. Maybe I'm wrong? I feel like there used to be a "screenshot only this window" hotkey but AFAIK that was replaced with snipping.
By "safely" I mean getting a screenshot of only the current window. I can tightly control what appears in just one window by, e.g. opening an incognito window in Chrome. I can sort of be careful to not capture things I don't want by using a separate desktop but I'd rather just grab a single window if possible.
Ultimately these will probably be shipped off to our content folks anyway, so they can make any edits to the image I request. I'd just rather give them starting material that's as clean as possible.
Windows Terminal has a feature where ctrl-clicking a link opens it in your browser. Anyone know if there is a screen-reader-accessible or keyboard-driven way to do this?
We use Teleport at work, which regularly prompts me to visit a long-ass URL to log in. The only way to do that is to use NVDA's review cursor to find the "http", set a start marker, arrow by character past the URL (which ends in a UUID so that's always fun since I have to review past 36 characters for that), set an end marker and copy the link. Rinse/repeat for any link dumped to the terminal. Takes me 20 or so seconds when all is said and done and I wish there was a faster way.
Does anything like this exist? I regularly find myself typing the same Slack bot commands, same shell commands, etc. Looking for a Windows app with a global hotkey which, when pressed, pops up an autocomplete box with titles like "Lock gateway," "Lock dev host", "Strip ANSI codes", etc. Each title is associated with text which, when the title is selected, gets pasted in. The autocomplete box also has an item for adding new snippets.
It'd probably be easy enough to write but it seems like it ought to exist already.
Wanted to gut-check a project idea to see if anyone other than me would find value in it.
I have NVDA running across a handful of machines. I also, over the years, have done lots of subtle config changes, installed a handful of addons I prefer, etc.
Was thinking it'd be nice to have an NVDA sync service. Create an account, link your NVDA installations to it. Optionally mark specific config keys/addons to not be synced from specific installs, but otherwise keep everything synced everywhere else.
Add an ollama image recognition endpoint as an optional, or possibly limited-use, service addon for anyone wanting that capability but unable to run/host it locally. That'd probably be a paid feature, but config/addon sync is probably easy enough to do for free. Hell, maybe even train a chatbot with NVDA docs, GitHub issues, etc. and see if it can handle support requests.
Creating/hosting that all should be easy enough, sans the AI bits which would eventually cost but I could probably prototype something for free, though I'd eventually need to monetize it even if only to pay its own hosting costs if I'm ever unable to cover those myself. This feels like something I'd use, but I'd rather not spend the time if I'm the only one.
Never thought I'd be that guy raving about how great his company is, but...
Just got out of a nearly hour-long work meeting with the folks we're contracting with about audio describing the visual aspects of our town halls. Demos, presentations, I'll have my own personal audio describer in a shared Slack channel and huddle describing things out loud and answering my questions. We're also going to hash out presentation guidelines so folks doing demos do a few things to make everyone's lives easier, including watchers on phones where they can't see commands or screenshares.
Mind absolutely blown. Do FAAMG-scale companies do this for their employees, or just mine? This was a unique situation for everyone in the meeting, including the multiple companies contracting and subcontracting on this, so I suspect I just hit the jackpot and no one else does this.
I'm truly spoiled. Fly.io set the "cares about employees" bar pretty high and if I ever lose this job I'm going to be hard-pressed working for anyone else.
@matt@chris That's so wild. We do many of our meetings on Gather, this...let's just say less-than-accessible meeting space platform where you walk around between rooms, move in and out of conversations based on proximity, etc. I noticed the chat wasn't accessible at the time (they may have fixed this since) so after a week of me being there they just moved all our meeting chats to Slack. I figured there'd be some pushback or complaining, but no. Everyone just changed direction last June and all of our town hall chats are in Slack. We still use Gather for some things, but any larger meetings for my own team or that I'm distinctly interested in are on Zoom now, and no one has done anything to make me feel the slightest bit bad about having to change things, even implicitly.
I'm simultaneously surprised and not surprised that bigger companies don't do better.
@matt@chris Oh yeah, and we talked about using asciinema for some of our demos, our CEO even set up our own internal server so nothing sensitive has to leave our networks, but I don't think that got a lot of traction. I feel like that's probably due to just not pushing it more, though, since that's a change every presenter would have to make. Still though, I bet folks would do it if I asked, and we may start doing it where possible so the poor audio describer isn't having to read all the things. 😀
Set up an ollama server on my home network a while back and it's kind of magical to have all this capability just available to me anywhere (thanks to my Wireguard VPN, at least.) It's CPU-bound so a bit slow, but I've used it for everything from cleaning up writing to generating fanfic for my cat.
Since I set it up a couple months back, they've added support for image classifiers and OpenAI compatibility. Sorely tempted to buy an external GPU to speed things up. I do have access to cloud-hosted GPU hardware through work, but there's kind of a cool factor to running the AI on my assorted pile of hardware in my home. Wonder if anyone has integrated it with Home Assistant's voice assistants yet.
I haven't done a #fedora install in a while. Previously I'd boot into a live environment, manually start Orca, click a desktop icon called something like "Install Fedora" and go through the installation accessibly. Now the F38 Workstation spin seems to drop me straight into the installer and I can't start Orca with either alt-f2 or the keyboard shortcut.
Is this how the installer now works? If so, is there another method to go back to how it worked before?
If it's not obvious from the above, I'm a blind screen reader user. Right now I'm pressing Enter right when the boot media loads. If there's some combo of arrows/enter that can get me to the previous live environment, that'd be very helpful.
Having used it for a couple days, I like the 75% keyboard form factor. Mostly, I miss my Windows key on the right, and my applications key. However, I can now (probably) flash the board with custom QMK firmware. I'm going to change the right function key to Windows, and the slash to applications if it's held down. I'm also changing the three keys on the right to home, end, and something else I haven't thought of yet. I'll make combos of these three act as page up and down. The beauty of QMK.
@objectinspace@alexhall I really do wish Home and Delete weren't right beside each other, though. As a Thunderbird user, I can't count how many times I deleted some random message when what I wanted to do was hit Home and move to the top of my inbox.
Otherwise I like 96% but I'll probably be back to 100% if I can't figure this one out.
@objectinspace@alexhall OK, I'm probably going to regret asking this, but numpad delete? Not familiar with that one.
Do keep me posted if you do anything about this. I think I saw you posting about the Keychron K4 a bit ago and I didn't want to be that nerd, but that's what I have too. Had it for a bit over a year and still happy with it, delete key position not withstanding of course. 😀 I'm pretty good at hitting F12 because my other fingers have the backspace key as a guide, bit it's definitely a tricky keyboard to get used to.
Does anyone know of a CLI utility that can take a bunch of logs, parameters like which log level to focus on and maybe a few others, pipe that to an LLM, and get back an executive summary of any problems found, trends, potential issues, etc.?
As a screen reader user, I can't stand having to read logs because they're in about the least speech-friendly format ever (first the timestamp, next the level, then finally the thing I care about, rinse/repeat for each and every line.) Didn't occur to me until recently that an LLM might be a good solution for dumping in a bunch of logs and getting back some succinct analysis.
Asking because if it doesn't exist, I might score some work time to build it. But I'd rather not if someone already has.