Using GPT-4 to generate 100 words consumes up to 3 bottles of water — AI data centers also raise power and water bills for nearby residents | Tom's Hardware tomshardware.com/tech-industry…
Can you imagine a city with less noise and less air pollution?
Today we experienced it here in Brussels and in hundreds of EU cities on #CarFreeDay!
1929 cities from 43 countries are participating with a single motto: ‘Shared public spaces’ and show their commitment to cleaner and more sustainable urban transport during #MobilityWeek.
The BT Speak (blazietech.com/bt-speak-pro), particularly in its "traditional" (non-GUI) mode, takes such an interesting approach to implementing a UI specifically for blind people.
From the mid 2000s until I got my BT Speak, I thought the only viable options were to either implement a screen reader for a GUI, or implement a fully custom UI designed specifically for speech output and whatever kind of keyboard or keypad the device supports. The BT Speak does neither. 1/?
Instead, it builds on the Linux text console and BRLTTY (brltty.app/). BRLTTY is best known for supporting Braille displays, but it also supports speech output, as well as input from a stand-alone Braille keyboard. The BT Speak uses it this way. BRLTTY's maintainer, Dave Mielke, is one of the main developers of the BT Speak software. 2/?
By the way, all the source code for the BT Speak software is right there on the device. There's no attempt to obfuscate the code or to prevent device owners from accessing it. It's not under an open-source license, and the GitHub repository is private. So out of respect for Blazie's ownership of the code and their wish not to share it publicly, I won't share any code. But I didn't sign an NDA, so I can still write about what I've learned while studying it. 3/?
The BT Speak's central application is an editor, a modified version of GNU nano. Now, nano is under the GPL, so I guess I'd be entitled to share *that* code if I wished.
The other key component is what looks like a home-made terminal multiplexer, similar in function to screen or tmux, but tailored for the BT Speak UI. So, according to the ps command (which I can run while accessing the device via SSH), that runs on tty1, and the editor runs on pts/0. 4/?
OK, it looks like the program that owns tty1, called ui-main, isn't a general terminal multiplexer like screen or tmux. Unless I'm missing something in that program or the libraries it depends on, it only ever runs the editor in a pseudo-terminal. The editor also seems to be the only interactive application that can be backgrounded by bringing up the options menu. So in a way, all other applications are secondary, and multitasking is limited. 6/?
Those other applications (calculator, calendar, contacts, audio player, games, etc.) are accessed through the options menu, which is implemented in ui-main. It looks like the menus are implemented using ncurses, in a way that's designed to work well with BRLTTY's speech output. To back out of a menu, you press the Z chord on the Braille keyboard (dots 1, 3, 5, 6 plus the space bar), which maps to Escape in the terminal. 7/?
That use of the escape key is one place where the fact that we're using an emulated VT100 leaks through. Since the escape character (0x1b) is also the prefix for some other VT100 keycodes, as well as keystrokes that use the Alt modifier, a library like ncurses has to wait a bit before interpreting that character as an Escape keypress. So that command feels laggy. 8/?
So when you run an application from the menu, it looks like ui-main relinquishes control of the terminal to the application. This means, among other things, that all the applications have to directly support the convention of using Escape (Z-chord) to exit the application and return to the menu. I've verified, for instance, that if I run the calculator application in an SSH session, pressing Escape in that SSH session exits the application. 9/?
The calculator is perhaps the most classically Unixy of all the BT Speak applications. It's a shell script that runs GNU bc as a subprocess and interacts with it through pipes. The shell script includes a .sh file that defines some functions that are reused across multiple shell scripts. TO prompt for the expression, the calculator shell script calls one of these shell functions, which calls a Python script... 10/?
The Python script uses a "dialogs" Python module, which is also directly used by some other programs that are written fully in Python. The BT Speak system is quite polyglot. ui-main and the libraries it uses are in C. As I said, there are several shell scripts. And there's another program I'll get to later which is in Tcl.
I'll write more about this dialogs module after I've studied it. 11/?
Wow, the first thing I notice about this dialogs module is that its clearScreen function spawns the "clear" command as a subprocess. The whole system spawns processes like they don't cost anything. And yeah, I guess executing a C program on Linux costs very little. Still surprising though, and to be honest, I don't like it. 12/?
Of course, I should temper my criticism. What matters is that they shipped, while I only ever dreamed of developing an open pocket computer for blind people someday, and never came up with anything approaching a complete software design of my own, never mind the hardware. 13/?
The BTSpeak.dialogs Python module has a function, requestInput, that uses the curses module to prompt the user for a single line of text and process the input, with some basic editing. The prompt is shown in reverse video; I presume this is necessary to make BRLTTY speak it. The function somehow implements input in literary Braille. Note that BRLTTY has already translated the Braille keystrokes to characters by the time this function reads them from the terminal. Seems a bit roundabout. 14/?
To my surprise, though, most of the other functions in this dialogs Python module don't use ncurses directly as requestInput does. Instead, they invoke shell scripts which call whiptail to display dialogs. Yes, full TUI dialogs with border decorations and all. There seem to be some tweaks to the display for BRLTTY's benefit, though. 15/?
I think I would have at least had the Python module call whiptail directly, if not use the Python newt module (newt is the C library that whiptail uses). Going through shell scripts adds another layer of formatting and parsing code, which seems needlessly error-prone. 16/?
BTW, the source code on the BT Speak is a full checkout of the private Git repository, with all history. Dave Mielke, the BRLTTY maintainer as I mentioned earlier, is the top committer by far. He's clearly a master of the traditional Unix way, that is, lots of small programs written in multiple languages. 17/?
@mcourcel You can bring up a shell, either through the menus or by pressing all the Braille keys at once including Space (yes, seriously), and from there get to Python. Once you're in the shell though, you have to be familiar with BRLTTY's screen review commands, as BRLTTY doesn't automatically speak all output like, say, Speakup does.
The games seem to be a recent addition to the BT Speak, and these use another interesting hack, specifically, that Tcl program I teased earlier. BRLTTY doesn't automatically read all terminal output as other screen readers do. But the games that have been added so far are all line-oriented, and of course when playing, one doesn't want to be distracted with screen review commands. So there's a program called say-program-output... 18/?
say-program-output runs the specified program as a child process. The child process receives input from the terminal, but its output goes to the parent process, which directs that output to the text-to-speech engine. So, in effect, a very crude screen reader for line-oriented programs. 19/?
One downside, which I immediately noticed when I started playing Hangman on the BT Speak last night, is that there's a noticeable delay between when I type a character and when it's echoed back. I asked about this on the BT Speak support mailing list (just before I started this thread), and Dave says it's because say-program-output uses a tool called expect, which can only specify timeouts in whole seconds. I'm tempted to rewrite this program in Python using async I/O. 20/?
I could go deeper on various things that I've discussed on this thread. But I think I'll just briefly discuss the software update process and wrap up with some closing thoughts.
The update process pulls from the private GitHub repository (using an authentication token, presumably a read-only one, that can easily be found on the device), then does several steps (announcing each one), including installing Raspbian packages and compiling some things on the device itself. Then it reboots. 21/?
The update process might be the most user-visible contrast between the BT Speak and what we've come to expect from mainstream, polished, sealed (even if technically open-source) consumer platforms. With Android, for instance, the system image is a read-only filesystem, and the update process downloads that image onto an alternate partition then reboots into it. It's a definite case of what @glyph once compared to "fused silicon" (blog.glyph.im/2006/04/fused-si…). 22/?
In its architecture, if not its license or level of documentation, the BT Speak software fulfills the vision for end-user free software outlined in that old post by @glyph that I linked in my previous post. There's some C code, but basically all of the applications and system management utilities are written in interpreted languages, such that they can be run directly from source. And the update process is impressively transparent about what it's doing, which might spark curiosity. 23/?
@glyph I'm aware I tagged you in the middle (hopefully towards the end) of a mega-thread. I hope it wasn't too confusing seeing that post out of context.
The founder of the company behind the BT Speak, Deane Blazie, originally developed a PDA for blind people called the Braille 'n Speak in the 80s. (We often call that class of device a note-taker because of its original killer application.) Out of necessity given the hardware, the software for that device was much more monolithic, and it was closed-source. Still, it included a BASIC interpreter, and that was enough to help at least one blind person I know get started learning to program. 24/?
I hope that many blind kids have BT Speak units bought for them in the years to come. It seems to me that the openness of the software in this device (in practice, if not in license) could help future generations of blind hackers get their start. 25/25
(At least, I think this is the end of the thread.)
OK, I guess I'm not quite done after all. Having waxed positive about the BT Speak's openness, and acknowledging again that their real, shipped product is worth more than my criticism, I do wish they had written the whole UI and application suite in Python. The current mix of C, shell scripts, and Python is an inconsistent mess, albeit a functioning one.
I just wanted to express my appreciation for your taking the time to write this whole thread. This is a good perspective and is simple enough for those of us who don't really understand this stuff to get it. Thank you.
@techsinger No problem. As I was writing it, I thought I should really revive my blog. But it's easier to just dive in and do an off-the-cuff thread like this than to produce a polished, edited long-form blog post.
@techsinger I also wasn't actually trying to tailor my explanation for non-programmers. I think I was more conscious of trying to make it understandable to programmers who aren't immersed in access technology for blind people.
@techsinger Agreed! I really wanted to know what was driving this note taker. If they ever throw in a braille display into it, I might consider purchasing one. Especially if I can create my own programs, or run other Linux apps..
@mcourcel @techsinger To be clear, the only officially supported way to run third-party Linux apps is in the GUI ("desktop") mode, with the Orca screen reader. In traditional mode, you could open the shell and run CLI/terminal apps with BRLTTY. And in theory, you could edit the menu files to add third-party apps to the menus. But any edits to the menus, or other code in the BTSpeak tree, would risk causing conflicts on the next system upgrade.
@miki The origin Git remote is configured to pull from GitHub via HTTPS, with a token embedded in the URL in .git/config. No, I won't share the token. I presume it's read-only but don't wish to test that.
Come to think of it, this is actually not a great idea.
If they need / are forced to ever revoke that token for any reason, if it expires, or if Github changes their authentication mechanism, the devices are pretty much FUBAR and will need a far more arduous process to make them updatable again.
Fascinating thread. Not a programmer but like to read technical things anyway. Would like one of these units but can't justify purchase. Still, hope to see one one day and then make a decision.
Very interesting writeup on this. Now if they want to keep everything open-source, I don't really see how they can allow NLS talking book playback. Yes I know that in a good security system only the private keys and not the source code need to stay hidden, but still I can't see NLS being happy with anyone being able to look at the source code on-device and find out where those keys are stored.
In der Anwendung von freier, offener und wirtschaftlich fair betriebener Software sehen die Lukis eine praktische Verbindung zur #Freiheit des Evangeliums und vieler Narrative biblischer Texte.
Das war auch an diesem @luki@kirche.social Wochenende im Theologischen Seminar Herborn wieder zu spüren.
I've got some news about office environments. The absolute worst part for me was dealing with the rampant gossip and laziness. People would chat all day, accomplish next to nothing, and then pretend they were some kind of heroes for staying late. People walk around, take lunch or bathroom breaks, attend meetings, etc. Office or Home people will walk around. Stop writing stupid stories to promote back-to-office culture and risk everyone’s health.
All the session videos will be online in the next ~week or so, and one of our speakers will be generously mirroring them on PeerTube here: video.innovationhub-act.org/c/…
And we look forward to rolling out our own official PeerTube channel before MatrixConf 2025!
Does anybody have any documentation or knowledge about how Apple does certificate revocation (valid.apple.com), besides the WWDC 2017 presentation that very briefly summarized it?
My understanding: they use a CRLite-like Bloom filter whose positives trigger live OCSP lookups when stapled OCSP responses don’t exist, and this filter covers all revocations (like CRLite). This seems to check out with API docs and a cursory look at the DNS logs of the devices I’m testing with.
I’m especially interested given that Apple (along with all the other voting browser vendors) voted for a CA/B ballot that made OCSP optional and required CRLs/sharded-CRLs, yet Apple heavily relies on OCSP. What are they doing to reduce their reliance on OCSP?
@jan Já používám Pixely, už pár let, na jiné telefony nejde nahrát #grapheneos, ale Pixel 6a má baterii fakt nic moc a to nemluvím o tom, že se občas i dost zahřívá.
@cynik_obecny @sazy Já bych ho klidně prubnul, už jsem o něm také uvažoval. Jenže musím u auta vyměnit olej a koupit nové zimní pneu. To je pořád něco ))
Reading about how social media collapses contexts so Black Women in particular feel compelled to put forth a lowest common denominator "respectable" self to real and imagined white audiences. There is little room for code switching to be able to present your authentic self when spaces aren't created with you in mind. academic.oup.com/jcmc/article/…
As some of you may already know, System76 is working on their new Linux graphical interface, the COSMIC desktop. They have created a form with some questions related to accessibility. If anyone is interested in participating in the survey, please access the address below:
"Certificate authorities and browser makers are planning to end the use of WHOIS data verifying domain ownership following a report that demonstrated how threat actors could abuse the process to obtain fraudulently issued TLS certificates": Google calls for halting use of WHOIS for TLS domain verifications arstechnica.com/security/2024/…
WHOIS for verifying domain ownership in 2024. I’m amazed that they’re still using an unencrypted and unauthenticated protocol for this. It’s a shame RDAP support isn’t universal yet.
one of my regrets is when i wrote that article on stylometric fingerprinting. i should have waited to get better at stylometric mimicry and focused on how i did that to improve anonymity and make it harder to connect alts (a bunch of custom vale.sh rules to remove style and conform to a boring technical style, then reintroducing a target’s style). instead i kind of spawned this community of people interested in using it for forensics.
i know they’ll see this post but idc at this point lmao
on the other hand, LLM slop awareness has created more bad stylometric analysts than any blog post of mine ever could.
“this was clearly written by an LLM because it uses [very common vocabulary preferences]!”
There are clear tells, and LLMs like ChatGPT do have a set of (cringeworthy) emergent loose style guides, but I usually don’t rely on this sort of thing. I don’t have a linguistics background. I can claim to be good at fingerprinting avoidance but actual fingerprinting is too error prone.
My BT Speak Pro (blazietech.com/bt-speak-pro), a pocket computer designed for blind people using the Raspberry Pi Compute Module 4, consistently has an internal CPU temperature of around 50 degrees Celsius (as reported by "vcgencmd measure_temp") when mostly idle (that is, all that's happening is the SSH connection from my PC where I'm running that command). I wonder if that's normal for the Pi 4, and if it can be lowered, or if that's just a limitation of that SoC.
Now I'm curious about the CPU temperature of my phone when it's idle, when it's locked but playing audio, when I'm actively using the UI, etc. My current phone is an iPhone though, so I probably can't access those stats. Thinking about switching back to Android though.
Without active cooling, in the form of a fan of some sort, that is around average for an idle Pi4. But I'm not sure if BT Speak has any fans or not. Plus, also not sure what it might be doing in the background, aside from serving your SSH connection. They definitely run hot.
I do regularly, from the built in temp check via a voice command. it's available somewhere in the menus, but I do it the lazy way since i'm always on wifi,.
@scruss How can you reduce thermal activity? I'm happy to experiment and find out what works and what doesn't. I know about the powersave CPU frequency governor. Anything else?
Seriously, I'd check with Blazie if they think this could cause problems. You're not hitting thermal problems at the moment, and you might find the device becomes slow or unresponsive
@scruss I don't actually depend on this device for anything important; I bought it specifically to experiment with it. So as long as I don't brick it and can undo my changes, I'm happy to try things that might degrade performance, especially if they can reduce power consumption (which I assume is correlated with heat) and prolong battery life.
“You are not special. You're not a beautiful and unique snowflake. You're the same decaying organic matter as everything else. We're all part of the same compost heap. We're all singing, all dancing crap of the world.”
When people call you a “snowflake” just remember they're quoting Fight Club.
Fight Club.
A satire written by a gay man about how male fragility causes men to destroy themselves, resent society, and become radicalized. And that Tyler Durden isn't the hero but a personification of the main character's mental illness.
His “snowflake” speech is a dig at how fascists use dehumanizing language to breed loyalty from insecure people.
Do you have access to the most effective GOTV tool in existence? You do if you live in one of these battleground states or districts! We’ll connect you with 10 key voters in your neighborhood that we need to reach to win in November. It’s really that easy to make a difference in this election! Get started today: indivisibleproject.formstack.c…
Fietsberaad is working on intergating their cycle park data with OSM. Join us on Oct 17th at their offices in Brussels to get to know them and work on the project. Drinks & food are on them! Register at meetup.com/openstreetmap-belgi…
Happy Software Freedom Day! Thank you to all who contribute to free and open source software around the world! #SoftwareFreedomDay2024 #OpenSource #FOSS
Google have really stolen an accessibility march on Apple with image descriptions with rich detail, built right into Talkback. No more sharing to other apps, just focus on an image, single tap with three fingers and select Describe Image. Browsing images on social media has never been so interesting or easy. Here is an example, taken at Bewdly safari park:
A rhinoceros is standing in a grassy field in front of a wooden fence. The rhinoceros is facing the left of the image, its head is down and its long horn is visible. The rhinoceros is gray in color with a large, bulbous body and short, thick legs. The grass in the field is a vibrant green. The fence is made of brown wood and has a wire mesh on top. Beyond the fence, there are trees with green foliage. The image is taken from a slightly elevated angle. The time is 11:29. Thank you Google. As a totally blind person from birth, I never thought I would be interested in pictures, I love them now.
This thread highlights one of the things I don’t like the most about people on the fedi: the insufferable arrogance of people (usually tech folks) who think this place should stick to their narrow vision and openly disparage anything that deviates from from that.
The beauty of the fedi is we can literally do anything we want here. It doesn’t have to be one way or the other. There is enough space on the internet for us to have our own special little places.
The dogma I see from people complaining about not wanting ads, when you are literally promoting stuff in your bio or not wanting influencers when we boost the sentiments we agree with and saying everything should be free when being on the fedi costs money is so ridiculous it’s beginning to border on straight up bigotry. It’s ok if one doesn’t like these things, but fighting to keep them out of the fedi because of that is replicating the centralized environments we say we allegedly one to get away from.
Not having these things is not what differentiates the fedi from centralized platforms. Giving people the tools to build whatever experience they want safely is what is going to put this place over the top.
Anything less than that and it’s just recreating less featured versions of stuff we already have.
Mekka does a good job of laying out why basing the direction of the fedi based on what we don’t like is a losing game, especially when those dislikes are based on shallow and myopic points.
My dislikes of places like TikTok outpace my likes, but one of the things it has over the fedi is diversity of content, which is ridiculous to me considering it’s a closed platform with wonky curation.
I can fun stuff, serious stuff, comforting and informative stuff all together because it has a variety of people from all walks of life sharing what they think is interesting.
The fedi has always fought against this diversity while claiming this place is ‘better’ while refusing to acknowledge the tooling is worse, safety is an after thought in most places and most fedi projects are managed by untrustworthy people.
I believe in the fedi becoming the de facto experience on the web, but there is a lot of to do to get there. We still haven’t realized the potential of this place because, quite honestly the platforms we have at the moment are not good enough for use by people who don’t have extensive experience in tech. And that’s not even getting into the inherent bigotry of the fedi for anything not white.
This place still just lacks the basics when it comes to building dynamic communities.
Just learned about Emissary, the social web toolkit.
At first glance, it looks like an extensible server that lets you define/extend ActivityPub types with your own, or with custom workflows/access rules.
Now I have to keep myself from going down a million rabbit holes with this. Tempted to dust off my old federated Steam/Itch game/digital project distribution idea, seems a whole lot more doable with something like Emissary than entirely from scratch.
))
Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •The BT Speak's central application is an editor, a modified version of GNU nano. Now, nano is under the GPL, so I guess I'd be entitled to share *that* code if I wished.
The other key component is what looks like a home-made terminal multiplexer, similar in function to screen or tmux, but tailored for the BT Speak UI. So, according to the ps command (which I can run while accessing the device via SSH), that runs on tty1, and the editor runs on pts/0. 4/?
Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •The Python script uses a "dialogs" Python module, which is also directly used by some other programs that are written fully in Python. The BT Speak system is quite polyglot. ui-main and the libraries it uses are in C. As I said, there are several shell scripts. And there's another program I'll get to later which is in Tcl.
I'll write more about this dialogs module after I've studied it. 11/?
Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Martin from Toronto
in reply to Matt Campbell • • •Matt Campbell
in reply to Martin from Toronto • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •I could go deeper on various things that I've discussed on this thread. But I think I'll just briefly discuss the software update process and wrap up with some closing thoughts.
The update process pulls from the private GitHub repository (using an authentication token, presumably a read-only one, that can easily be found on the device), then does several steps (announcing each one), including installing Raspbian packages and compiling some things on the device itself. Then it reboots. 21/?
Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Glyph
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •Matt Campbell
in reply to Matt Campbell • • •I hope that many blind kids have BT Speak units bought for them in the years to come. It seems to me that the openness of the software in this device (in practice, if not in license) could help future generations of blind hackers get their start. 25/25
(At least, I think this is the end of the thread.)
David Goldfield reshared this.
Matt Campbell
in reply to Matt Campbell • • •Jayson Smith
in reply to Matt Campbell • • •Tech Singer
in reply to Matt Campbell • • •Matt Campbell
in reply to Tech Singer • • •Matt Campbell
in reply to Matt Campbell • • •Tech Singer
in reply to Matt Campbell • • •Martin from Toronto
in reply to Tech Singer • • •Matt Campbell
in reply to Martin from Toronto • • •miki
in reply to Matt Campbell • • •Matt Campbell
in reply to miki • • •@miki The origin Git remote is configured to pull from GitHub via HTTPS, with a token embedded in the URL in .git/config. No, I won't share the token. I presume it's read-only but don't wish to test that.
The first line in /home/pi/.bash_history is:
/BTSpeak/Tools/prepare-for-shipping -a
Haven't gone looking at other dotfiles.
miki
in reply to Matt Campbell • • •Come to think of it, this is actually not a great idea.
If they need / are forced to ever revoke that token for any reason, if it expires, or if Github changes their authentication mechanism, the devices are pretty much FUBAR and will need a far more arduous process to make them updatable again.
Steve
in reply to Matt Campbell • • •Matt Campbell
in reply to Steve • • •Martin from Toronto
in reply to Matt Campbell • • •Matt Campbell
in reply to Martin from Toronto • • •miki
in reply to Matt Campbell • • •I've always wondered whether you could do something like that, but with Emacs and a hypothetical "Braille notetaker mode".
It seems like the perfect piece of software for that kind of device.
Andre Louis
in reply to Matt Campbell • • •Jayson Smith
in reply to Matt Campbell • • •Matt Campbell
in reply to Jayson Smith • • •