Die allermeisten „KI“-Bildbeschreibungen die ich bisher gelesen habe sind fast so nützlich wie der Hinweis „da ist ein unregelmäßig geformter, erdfarbener Fleck auf dem Gehweg vor dir“ an eine blinde Person die dabei ist in eine ungesicherte Baugrube zu fallen. Der „optische Eindruck“ stimmt, die relevante Information fehlt.
Spielt „Montagsmaler“ oder „Tabu“. Nehmt Texte etwa vom Altbot-Feed, versucht euch vorzustellen was auf dem beschriebenen Bild zu sehen sein könnte, und dann schaut euch das Bild an das der Bot beschreiben sollte.
„Ein älterer Mann im blauen Anzug sitzt mit verschränkten Händen am Schreibtisch“. Ob es Donald Trump, Friedrich Merz oder Onkel Günter ist weiß der Altbot nicht, und wenn es nicht im Originalpost steht erfährt es auch die Person nicht die auf den Alt-Text angewiesen ist.
„Das Bild zeigt einen Blick durch ein schwarzes Mesh-Raster, hinter dem sich ein geschlossener Bereich befindet. Im Inneren liegt eine große Menge weißer, weicher Material über dem Boden. Links ist ein Bereich eines Gebäudes mit einem hellbraunen Mauerwerk und einem dunkelfarbenen Struktur zu erkennen. Am unteren Rand befinden sich mehrere braune rechteckige Objekte, möglicherweise Holzbretter, die verteilt sind.“
Blick durch ein Fenster mit Insektenschutznetz auf Schnee auf einem Balkon.
Ein schneebedeckter Winterlandschaft zeigt ein Gewässer, vermutlich ein Fluss oder ein kleiner See, umgeben von schneebedeckten Boden. Bäume mit schneebedeckten Ästen, sowohl Evergreen- als auch Barentiere, winden sich am Ufer des Wassers und erstrecken sich in die背景.
Eine Sammlung von schneebedeckten Zweigen liegt am Wasser im Vordergrund. Ein kleines dunkles Gebäude ist teilweise sichtbar von der linken Seite des Bildes. Der chinesische Charakter “冬” (Winter) ist auf dem Baumstamm im oberen zentralen Bereich geschrieben.
Im Originalbild sind weder ein Gebäude noch chinesische Schriftzeichen zu sehen.
„Ein zugefrorener Teich im zugeschneiten Wald“ zu schreiben wäre natürlich Zeitverschwendung für den Autor gewesen. Der Alttext ist dagegen Zeitverschwendung für alle Menschen die darauf angewiesen sind.
Ich weiß, es gibt oft den Wunsch nach möglichst detaillierten Bildbeschreibungen. Aber doch bitte nicht diesen unleserlichen Textcluster an dem fast nichts stimmt.
„Das Foto zeigt eine Person, die einen schwarzen Mikrofon hält, dicht an ihrem Mund, mit einem dunklen Anzugrock, einem hellblauen Blusehemd und einem gepunktetem Hemd. Die Person hat kurze, dunkle Haare, die mit einer Seitenteilung geschnitten sind, und der Hintergrund ist dunkel mit einem verschwommenen, warmen Lichtquelle hinter ihr sichtbar.“
Man(n) trägt jetzt zwei Hemden. Tippe auf Balenciaga.
Wieviele Blinde haben Altbot wegen dieser Scheiße inzwischen stummgeschaltet oder blockiert?
Vielleicht war der anfangs tatsächlich mal brauchbar. Aber das Modell wurde mehrfach gewechselt, und wer das Ding jetzt noch nutzt ohne wenigstens mal ab und zu reinzulesen was es inzwischen produziert… ich weiß ja nicht. Ich hänge immer noch an den „sich windenden Evergreen- und Barentieren“ von vor ein paar Minuten fest.
Ich komme mir ja als Sehender schon von solchen Texten verarscht vor. Wäre ich blind hätte ich möglicherweise das Gefühl mir bietet jemand eine Lesebrille an.
Please stop using “poorly trained" when describing the murderous ICE agents. Rene Good's killer was a combat veteran and 10 year ICE veteran. It's not the training.
I can't complain. Some days are harder than others, but today is a pretty good day here. Especially after I woke up to no power this morning, but it came back right in time for work. woohoo!
Join the Internet Archive & partners for PROTECT OUR FUTURE MEMORY 🏛️
📚 This #webinar introduces the growing international movement to ensure libraries & other memory institutions have the same rights online they’ve long held in the physical world. 🌐
Libraries, archives, and museums have always collected, preserved, and provided access to knowledge for the public good. But in the digital age, those foundational rights are increasingly under threat. Join […]\n
The absolute *minimum* you can do right now is email your representatives in the Senate and Congress to tell them how you feel about what's going on right now. It'll take only 5-10 minutes of your time. Do it now.
I know a guy who used to be a conservative, until he went to college. It wasn't "woke professors" that change him to a liberal, it was being in an actual city instead of the small town he grew up in.
Meeting more diverse people made him realize he was fed lies about others while growing up.
So yeah, if you ever wondered why the right wants to shut down education and universities, that's one reason.
POLAND: Siemens Mobility and Newag have signed a memorandum of understanding to strengthen their co-operation in the Polish rail market and expand their previous joint work on metro trains into the high speed rail segment.
In the context of enshittification, there is one law that seems to be a constant. I mentionned it here and here:
Every piece of software will evolve beyond its intended purpose to sell you something.
Austerity and neoliberalism mean these institutions should be destroyed so we can focus on boosting corporate profits without the “waste” of analysis or memory. cbc.ca/news/canada/ottawa/war-…
We knew this was coming, but now the clock is running. From Privacy International:
"Yesterday the Trump Administration announced a proposed change in policy for travellers to the U.S. It applies to the powers of data collection by the Customs and Border Police (CBP)."
"If the proposed changes are adopted after the 60-day consultation, then millions of travellers to the U.S. will be forced to use a U.S. government mobile phone app, submit their social media from the last five years and email addresses used in the last ten years, including of family members. They’re also proposing the collection of DNA."
PI linked to and summarized a Federal Register entry describing the proposed requirements:
-All visitors must submit ‘their social media from the last 5 years’
-ESTA (Electronic System for Travel Authorization) applications will include ‘high value data fields’, ‘when feasible’ ‘telephone numbers used in the last five years’ -‘email addresses used in the last ten years’ -‘family number telephone numbers (sic) used in the last five years’ -biometrics – face, fingerprint, DNA, and iris -business telephone numbers used in the last five years -business email addresses used in the last ten years.
1. Other countries may, and I think probably, will impose similar requirements on US people trying to enter those counties. In other words, the US would approach becoming a closed kingdom, like N. Korea.
2. Just wait until someone remembers Little Bobby Tables (image below).
The US have been fingerprinting for over twenty years. Each time I complained about it, Canadians and Americans looked at me like "what are you talking about". And no reciprocity either because Europe is a bunch of cowards.
I feel for anyone in the travel, tourism and hospitality industries, which make up ~ 10M jobs and ~ 3 percent of the nation's GDP. From the U.S. International Trade Administration (trade.gov)
"Inbound international travel to the United States plays a vital role in the Nation’s economy and promotes cultural exchange and understanding. Travel and tourism is the largest single services export for the United States, accounting for 22 percent of the country’s services exports and 7 percent of all exports in 2023. The travel and tourism industry contributed $2.3 trillion to the U.S. economy in 2022 (2.97 percent of the country’s GDP), supporting 9.5 million jobs."
The US have been fingerprinting for over twenty years. Each time I complained about it, Canadians and Americans looked at me like "what are you talking about". And no reciprocity either because Europe is a bunch of cowards.
I need a new Apple Watch. I think I'm going to go for the SE 3, since it's cheaper, and see what happens this fall. If I really want the Series 12, I have a trade-in. If not, I have an SE 3 I can keep for a while. It only lacks ECG, blood oxygen, and hypertension detection, none of which I need or regularly use. The Series 11 is thinner, but I'm used to the Series 9, so I'm not losing anything there either.
Not sure how much this is something you care about, but the SE 3 has a smaller screen and as a result Apple chose not to add the onscreen keyboard to it. So for text entry you’ll be limited to scribble, dictation or the phone keyboard feature when it decides to work.
@pitermach Thanks, I didn't realize that. I find I use dictation on my Series 9 anyway, so that's no great loss. Still, I didn't realize the SE lacked the keyboard.
Yeah they don't really advertise this. That was one of the reasons I upgraded from series 5 to 10. Payed off even more when they added Polish language to it. Where it really helps is when typing in passwords and such where dictation really doesn't work well.
Heading to @chaoss ConEU before FOSDEM? Be sure to catch our Lunch and Learn this Thursday Jan 29 at 13:00 with @linuxflower and Wayne Mery talking about what we're doing with Thunderbird community metrics and how can we do it better! 📈
As soon as I typed that sentence, I just knew that someone else had already typed it. So I checked, the old way. Sure enough, James Rendel had already opened a LinkedIn post with exactly that line. It is on his Substack. 1/11
"An entrepreneur who produces low quality, derivative, or mass generated output especially AI generated content typically prioritising speed, volume, or short term gain over originality, craft, or long term value." 3/11
This is the latest version of the same old get rich quick scheme. The tiny classified ads from Don Lapre, Tom Vu on the yacht, those late night seminars about flipping houses with zero down. 4/11
Startups will convince them that they can be writers, artists, app developers all with zero effort, even while they sleep. Anyone with an interest in business knows how this goes. 6/11
Low investment, low effort, low skill work will never survive against competition with more money and skill and who put in real work. However, there is a lot of money to be made selling this kind of dream. 7/11
I love conversational interfaces, and I am a big proponent of automation. The trouble is when these tools are not sold as productivity software to empower people who are doing work, with skill and effort, but as ways to create value without skill and effort. 8/11
Since much of our economy, especially in new technology, is driven by investors, the spectacular growth of startups serving the bottomless slopreneur economy diverts investment away from companies creating more durable businesses. 9/11
Metrics of scale take attention away from substance, diverting capital from ventures that create real jobs and build sustainable revenues, to those that build roach motels with impressive ROAS but dismal LTV. 10/11
If you are looking for a messenger that doesn't depend on US infrastructure (or any other country actually) take a look at #DeltaChat, an #European alternative messenger, it is #anonymous, #decentralized, end-to-end #encrypted, #openSource and free!
it is multi-platform, get it for your system at get.delta.chat
there is nothing misleading, meaning of sovereign: "Self-governing; independent."
with Delta Chat you can self-host so you are independent, and rule over your server, you can even self-host in a local network/mesh without even needing internet
as from the point of view of a country this also mean they can have their owm servers in their land and be sovereign as a country
Remember the first thing they did was to axe fiscal measure that would have increased revenue (but it was taxing the rich). Instead they make regular people pay through service cuts.
@hub Not my ideal, but for email right now I'm going to Proton. I'm keeping my eyes open for a better solution, but I'm tired of giving Google money. Reading about the ICE concentration camps a.k.a. "tender age shelters" where they keep hundreds of babies, toddlers, and kids under 12 years of age to use as leverage against both documented and undocumented migrants broke me yesterday, and Google is super fucking complicit in all of it. All the technofascists are.
“An insane encapsulation:🚨”The same day Trump’s agents kill a VA ICU nurse in broad daylight, the CEOs of Apple & Amazon join Trump to watch a private screening of a film about his wife directed by an alleged sexual abuser *was seen shirtless in the Epstein Files.” -P Thakker
🔎 Wie schreibt man eigentlich eine gute, barrierefreie Bildbeschreibung?
Barrierefreie Bildbeschreibungen helfen Menschen mit Sehbehinderungen, Inhalte gleichberechtigt wahrzunehmen. Stelle dir Fragen wie: Wer/was ist zu sehen, was passiert, wo passiert es und welche Details sind für den Inhalt des Posts wirklich relevant. Wertungen („schön“, „hässlich“) sind weniger wichtig als neutrale Infos.
Das ist ein Post über gute Alternativtexte mit einem schlechten Alternativtext. Wenn ich so was sehe, frage ich mich, ob solche Posts geschrieben werden, um gut auszusehen? Der Satz im Alttext ist ja nicht mal grammatisch richtig. Wird nicht damit gerechnet, dass den jemand liest?
Ich bin blind, Übersetzerin und Master BK. Und wunder mich.
Local AI friends: I want to get a good GPU for running local AI models. This custom-built tower currently has an NVIDIA GTX 1050 Ti with 4 gigs of RAM, which is hardly enough to do anything. I'd like 12-16 GB of VRAM, whatever would be better for running things like Whisper and coding models locally. My price range is about $350-$450, but I'm willing to push it up to about $500 for a good GPU that'll last me a while. Suggestions appreciated!
Good timing. Yup I'd need an upgrade as well, currently having an RTX 30-60 with 8GB, you can do stuff with it, but not a lot. But I think this is a bigger deal considering the prices and the fact I had to upgrade the case and power supply last time since the GPU required space and power, so once prices are doable again I'll probably get a bunch of hardware again.
@MewProjects hi. Trying to get fastSM up and running. When I try to register my mastodon account I’m getting an error about cannot complete request. In the error log I see something about not being able to register a device. Any idea what could be going wrong?
I can confirm that MSTDN Canada has a user agent filter list. Mastodon.Py is known to be blocked. Uncertain of what FastSM is sending for its user agent, but just letting you all know of a potential reason for the problems being experienced.
same bug i gave you last week. when you open a new account the web page fails to load to be authorized. i am assuming we are in fact talking about the same issue and i am fairly sure we are.
for what ever reason the web page will not open. it does in your browser though without issue. and by the way choosing the browser you want to use might just be a good idea rather than the default.
now if there was a way to copy the token generated in to the field that was expected i might actually get it to work. thanks for all your help it is appreciated.
the error is still the same someone not you perhaps is expecting something that isn't actually there. "Error registering app: Could not complete request: Expecting value: line 1 column 1 (char 0)
@gtbray So basically, the auth flow goes like this. You enter your server name, the app goes out to that server, registers itself as an app, then it does the whole authorization flow. It seems like in your case, the app registration part is what's freaking it out.
@gtbray If possible, I would like to request the application log a bit more. The only thing I see in the error log when this happens is something that says: unable to set default device, continuing.
@gtbray Hi. I just heard back from the instance administrator. He says that the application is still using the mastodon py user agent and that’s why it’s getting rejected. He sent me a screen showing what he’s seeing. I’ll try and get the text out of that.
@gtbray I found the problem. You are setting the user agent in the authorization process but not the registration process. I added it to the registration process in my source version and it’s authorizing now.
@gtbray The text of the log entry that the admin sent me is in the GitHub issue that the admin posted. He was definitely a bit more harsh than I would’ve been. However, it does look like the user agent is not fixed.
Overall, the snow probably made me more depressed than happy though. Goalball tournament got canceled, youth group which would've been the alternative, canceled. Just sitting here and at this point... I would have projects to work on, but at this point I feel, misplaced? I'm not yet really at the point that I could express or understand my current emotions, but I feel like I have to move on somehow, I'm still thinking about stuff that's just, in the past or no longer meant to be while I have not the slightest idea about my future. All I know is that I'm not satisfied with it.
@Aryan The main concept is that you have Brawlers, basically characters with different abilities, attacks, health etc. These Brawlers do also have gadgets, super powers, super attacks, ranks etc. You collect trophies by fighting with these brawlers, in various modes, either showdown (solo or duos) but also 3vs3 in modes like brawlball (football) or basketbrawl (basketball), gem grab, where the teams have to collect more gems than the other team while fighting each other of course, and when a player dies, they drop all the collected gems, hot zone, and a bunch more. Of course you have all the shit you'd expect around it: friends you can invite to play and form teams, clubs with rewards, quests which give you coins which unlock rewards on a reward track, brawler upgrades, brawler skins etc.
@Aryan To name a few abilities, you either have generic brawlers which just shoot bullets, but there are quite exotic ones as well like the projectile jumping from brawler to brawler, poisoning enemies while healing allies with the projectile, some that fly/jump, stun you etc etc. Of course some are slower, faster, so it's also a part to be strategic, for example a brawler with a wide range won't make a lot of sense when you play on a map with a bunch of walls since it makes more damage the farer it's away.
Went to fill out a W-9 this morning and realized Firefox's PDF form filling was...somewhat less than accessible. An hour or so later and I now have a Claude skill to do this for me. It finds and downloads forms, fills out details from natural language plus clarifying questions, applies image signatures using handwriting fonts, and does a visual verification pass on each page to make sure everything looks OK. I'm of course not confident enough yet to let this fill forms without human verification, and it doesn't yet handle non-fillable forms, but just being able to say "Help me fill out this form and sign/date it for me, then give me a basic sanity check" is a gamechanger even if I still have a sighted human in the loop. Reduces the process from lots of hand-holding and inaccessible editing, to a final and much quicker visual pass.
I've wanted an app to do this for over a decade, but wasn't super thrilled about writing something whose only value add was that it was the accessible version of something sighted folks take for granted. Now I have it, and it's less than 200 lines of markdown.
That's impressive. I still haven't messed with any of these AI tools; I should probably try to learn how to use them some time. I feel vaguely nervous that they are going to do something spectacularly wrong, but probably they can do useful things as long as humans are supervising them / checking their work.
Heard, I spend a lot of time watching mine. It almost never runs unsupervised unless I have it in a container, and I generally only leave it alone if it's working through a longer implementation plan which I've read, or if it's doing something like research. Even so, the fact that it works much faster than me is good, even if it does mean I spend more time checking its work later.
After I saw your post, I was asked to sign a form that had a canvas for me to draw a signature with the mouse, and now I'm wondering if Claude could help me with that. Despite what I said in my last comment, this is probably a case where I don't really care if the AI messes it up somehow. In this case, I pointed out that I can't sign the form without a sighted person helping me (I think I have an easier time advocating when I know that several other blind people are going to be affected, somehow), and they made the signature box optional, so it doesn't matter for now, but I've run into these things from time to time and don't know of a way to handle them and end up asking my roommate to help me.
Possibly. For that you want to look into MCP servers--specifically probably something called the Playwright MCP. This lets you use something like Claude Code to control a browser. It can either control your actual browser, or a sandboxed/headless browser. Sometimes I'll use this for a high-level sanity-check--have Claude implement something then ask it to use the code via browser and make sure it works.
Mal etwas zu #Inclusion: Ich werde belehrend kritisiert, weil ich - wie 70 % aller im Fediverse - #AltTexte nicht oder verkürzt schreibe. Ich teile ausdrücklich die Position, dass wir inclusiv agieren sollten. Aber manche Bilder dienen nur als Blickfang für Sehende, um Aufmerksamkeit für den Text zu generieren. Für eine Bildbeschreibung wie früher in der Schule fehlt mir meist die Zeit. Und vor allem: Es wäre möglich, dass Mastodon diese Bildbeschreibungen per Klick und KI generiert.
"dienen nur als Blickfang". Hast du es tatsächlich nötig, deinen Texten einen "Blickfang" hinzuzufügen?
"mir fehlt die Zeit". Aha. Verfassen einer Bildbeschreibung dauert etwa 2 Min. Wieviele Bilder pro Tag postest du?
Und dann noch "wie 70 % aller im Fediverse" – woher stammt diese Statistik? Dient als Begründung etwa so gut, wie "ich prügle mit, weil andere ja auch prügeln".
Hab gestern hier irgendwo gelesen, dass die Rate an Alttext im Fediverse auf 28% (oder so ungefähr?) gesunken ist, weiß aber die Quelle nicht mehr und ob das eine gute Quelle war.
Außerdem ist das eine schlechte Nachricht!! Die Heuchelei von wegen "Ich bin für inklusiveres Agieren, aber ich mach's selber nicht" find ich echt nicht ok.
Und wer selber sieht und ein gutes Internet hat, kann beurteilen, ob ein Bild "nur ein Blickfang" ist oder nicht. Wer blind/sehbehindert ist, kann aber nicht wissen, welche wichtigen oder unwichtigen Infos das Bild enthält und welche Infos damit diesen Personen vorenthalten werden!! Darauf wurde Herr Polenz auch schon hingewiesen, aber das Argument ignoriert er natürlich.
Die Betroffenen dermaßen auszuschließen und Bildbeschreibungen mit "wie früher in der Schule" zu verunglimpfen, ist echt niveaulos. Aber dann immer schön "Ich bin für Inklusion" behaupten...
Remember all this shit american software that nag you with question and never have a "No" answer? It seems that Americans can't take no for an answer, and like a bully threaten when shown resistance.
And I expect David "Chamberlain" Carney to bend backwards to spend more of this austerity budget to buy these flying bricks.
The US is threatening #Canada (again) if we don't but their F-35s. Here's what Denmark said about the ones they bought, and regret buying:
"They're in for repairs about half the time or even more," he said, "so the Americans have all the power of actually destroying our air force just by shutting down [parts] supplies."
Holger Dittmann
in reply to Holger Dittmann • • •Spielt „Montagsmaler“ oder „Tabu“. Nehmt Texte etwa vom Altbot-Feed, versucht euch vorzustellen was auf dem beschriebenen Bild zu sehen sein könnte, und dann schaut euch das Bild an das der Bot beschreiben sollte.
„Ein älterer Mann im blauen Anzug sitzt mit verschränkten Händen am Schreibtisch“. Ob es Donald Trump, Friedrich Merz oder Onkel Günter ist weiß der Altbot nicht, und wenn es nicht im Originalpost steht erfährt es auch die Person nicht die auf den Alt-Text angewiesen ist.
Holger Dittmann
in reply to Holger Dittmann • • •„Das Bild zeigt einen Blick durch ein schwarzes Mesh-Raster, hinter dem sich ein geschlossener Bereich befindet. Im Inneren liegt eine große Menge weißer, weicher Material über dem Boden. Links ist ein Bereich eines Gebäudes mit einem hellbraunen Mauerwerk und einem dunkelfarbenen Struktur zu erkennen. Am unteren Rand befinden sich mehrere braune rechteckige Objekte, möglicherweise Holzbretter, die verteilt sind.“
Blick durch ein Fenster mit Insektenschutznetz auf Schnee auf einem Balkon.
Holger Dittmann
in reply to Holger Dittmann • • •Ein schneebedeckter Winterlandschaft zeigt ein Gewässer, vermutlich ein Fluss oder ein kleiner See, umgeben von schneebedeckten Boden. Bäume mit schneebedeckten Ästen, sowohl Evergreen- als auch Barentiere, winden sich am Ufer des Wassers und erstrecken sich in die背景.
1/2
Holger Dittmann
in reply to Holger Dittmann • • •Eine Sammlung von schneebedeckten Zweigen liegt am Wasser im Vordergrund. Ein kleines dunkles Gebäude ist teilweise sichtbar von der linken Seite des Bildes. Der chinesische Charakter “冬” (Winter) ist auf dem Baumstamm im oberen zentralen Bereich geschrieben.
2/2
Holger Dittmann
in reply to Holger Dittmann • • •Im Originalbild sind weder ein Gebäude noch chinesische Schriftzeichen zu sehen.
„Ein zugefrorener Teich im zugeschneiten Wald“ zu schreiben wäre natürlich Zeitverschwendung für den Autor gewesen. Der Alttext ist dagegen Zeitverschwendung für alle Menschen die darauf angewiesen sind.
Ich weiß, es gibt oft den Wunsch nach möglichst detaillierten Bildbeschreibungen. Aber doch bitte nicht diesen unleserlichen Textcluster an dem fast nichts stimmt.
Holger Dittmann
in reply to Holger Dittmann • • •„Das Foto zeigt eine Person, die einen schwarzen Mikrofon hält, dicht an ihrem Mund, mit einem dunklen Anzugrock, einem hellblauen Blusehemd und einem gepunktetem Hemd. Die Person hat kurze, dunkle Haare, die mit einer Seitenteilung geschnitten sind, und der Hintergrund ist dunkel mit einem verschwommenen, warmen Lichtquelle hinter ihr sichtbar.“
Man(n) trägt jetzt zwei Hemden. Tippe auf Balenciaga.
Wieviele Blinde haben Altbot wegen dieser Scheiße inzwischen stummgeschaltet oder blockiert?
Holger Dittmann
in reply to Holger Dittmann • • •Vielleicht war der anfangs tatsächlich mal brauchbar. Aber das Modell wurde mehrfach gewechselt, und wer das Ding jetzt noch nutzt ohne wenigstens mal ab und zu reinzulesen was es inzwischen produziert… ich weiß ja nicht. Ich hänge immer noch an den „sich windenden Evergreen- und Barentieren“ von vor ein paar Minuten fest.
Ich komme mir ja als Sehender schon von solchen Texten verarscht vor. Wäre ich blind hätte ich möglicherweise das Gefühl mir bietet jemand eine Lesebrille an.