Writing image descriptions for AI slop is an exercise in recursion and arbitrary repetition
(see commentary in following posts)
![A cartoon drawing of a chicken with a Danish flag on its breast and EU flags on its wings, seated to the right of Donald Trump and JD Vance in a composition redolent of the February 2025 Oval Office meeting with Zelensky.
The Trump character has a speech bubble saying “We want Greenland and your eggs” and Vance’s says “Say thank you!”
The cartoon is displaying the AI generated alt text that accompanied its original posting, which says ‘Photo by Jake Broe on March 19, 2025. May be a cartoon of text that says “WE WANT GREENLAND AND YOUR EGGS. SAY THANK THANK YOU! YOU! [Arabic text] DENMARK.’](https://files.mastodon.social/media_attachments/files/114/188/125/343/904/219/original/4ceec58e489efe00.jpeg)
…Convenience wants us to replace actual care and discernment with a machine generated image description which:
- includes text that isn’t present but was in the enclosing post anyway
- is of a cartoon that “May be a cartoon” and isn’t a “photo”
- cannot recognise characterisations central to the image’s satire
- repeats words that are there
- and mistakes drawn elements for non-existent Arabic text
and to tell people who need image descriptions that this is ‘better’ than nothing
…And… convenience wants us to boost a shitty description without a thought or a care
…The Threads LLM (https://www.threads.net/@jakebroe/post/DHX4b1NOZlX) didn’t recognise
the characterisation of a chicken
or Trump
or Vance
or the setting that the “photo” was recalling
or the words as written
but apart from that this is better than nothing, right? Right? 
…We keep decrying a lack of care while taking *no care*.
Guess what happens
@urlyman Yeah Facebook's descriptions are neither use nor ornament. I've seen gems like "Maybe Pop art of text". Nothing about what the text even is.