#indieweb 2023-06-15
2023-06-15 UTC
plantroon, [tw2113_Slack_] and gRegorLove_ joined the channel
[Benjamin_Turne] joined the channel
# 
[Benjamin_Turne] ↩️ just seeing this, i’ll have to join next time
# 
gRegor Upcoming https://events.indieweb.org/tag/hwc
[schmarty], [lifeofpablo] and mandaris joined the channel; Mike[m]1 left the channel
# 
mandaris Hello! I've got two URLs https://mandarismoore.com/2023/06/14/this-is-a.html and https://mandarismoore.com/2023/06/14/testing-webmentions-one.html that I've been testing against https://mandarismoore.com/2023/06/14/testing-webmentions-one.html
# 
Loqi 📄 An article is a kind of post that typically has more structure than a simple note. Articles usually have a name (title), multiple paragraphs, and often subheadings, blockquotes, embedded images, and a footer of references or citations https://indieweb.org/article
# Loqi 📔 A note is a post that is typically short unstructured* plain text, written & posted quickly, that has its own permalink page https://indieweb.org/note
# Loqi A comment is a reply syndicated into the context of the original post https://indieweb.org/Reply
bartekzech2, chenghiz_, Xe and mandaris joined the channel
jbrr[m], win0err, plantroon, yar, gRegor, bterry, jeremycherfas, omz13, romangeeko, Seirdy, stokito[m], n8chz, jshmlr, [manton] and tei_ joined the channel
# tei_ does anyone know why it seems to not be possible to connect to IRC while a VPN is turned on? It's not a huge deal, but I have one connected by default and have to turn it off to get on
btrem joined the channel
# Allie tei_: Libera staffer here: If you check your IRC client's logs, the IRC server should tell you why you're unable to connect
# Allie (usually with VPNs it's because we require SASL auth from those ranges, to mitigate abuse)
# Allie libera.chat/guides/sasl
# tei_ thank you!
gRegor, tei_1 and tei_ joined the channel
# 
vikanezrimaya Anyone considered using AI image detection to automatically polyfill alt-text in case there is no alt-text on an image? Both in a microsub reader context, and for posting on your own website (if you forgot to fill the alt text for some reason)
# vikanezrimaya While AI may have its biases, having some alt-text may be better than having no alt-text at all. The only problem is — it probably would be better to run this locally, either inside of a client or on a server controlled by the service owner, instead of delegating this to the likes of sketchy companies like OpenAI
lawt joined the channel
# capjamesg Vika, I have lots of ideas here!
# capjamesg I am on a train but I shall respond in the next two hours.
# vikanezrimaya !tell capjamesg can't wait to read it!
tei_1 and [KevinMarks] joined the channel
# 
[KevinMarks] A lot of the ML generated alt text that I've seen is worse than useless
# [KevinMarks] I'd like to see better ones, but what I have seen in t he past are vague object recognition ones or keyword stuffed CLIP ones
tei_1 joined the channel
tei_1 and tei_ joined the channel
# gRegor https://www.instagram.com/alliancesandiego/ click on the first photo, that popover image has auto generated alt "May be a graphic of poster and text that says..."
# gRegor But the permalink for that photo doesn't have alt https://www.instagram.com/p/CspgkXKIPDt/
# gRegor From /Instagram looks like it was added to IG about 2018-11 https://www.engadget.com/2018-11-28-instagram-alt-text-visually-impaired.html and my earlier example I was looking at was before that, so that explains the simple [name] and [location] default alt.
btrem, Nuve, win0err, gaussianblue, tei_, tei_1 and [Murray] joined the channel
# 
[Murray] Relevant fun alt text/"AI" fun: https://adrianroselli.com/2022/08/ai-generated-images-from-ai-generated-alt-text.html
# [Murray] Personally I'd veer away from using ML to try and guess at alternative text for images. At best it might identify the _subject_ of the image correctly, but it will be a while before the _context_ can be inferred, and that's the important part. Just getting a list of descriptive words is quite annoying, and I imagine adding all of the "maybe, best guess" cruft is more frustrating (though I guess it can be funny at times too 😄)
# [Murray] I've heard of some assistive tech users running description-less images through ML models, and that feels like a better plan of action to me, because then it's within the user's control. So ideally add *actual* image descriptions, and if you forget or can't, then let the user try to generate it. At least that's a transparent system.