The discussion begins with JammyDodger expressing concern that the AI caption bot is malfunctioning, noting that captions were missing from images. Falco explains that the issue arose during LLM upgrades when the vision_enabled flag was accidentally left unset. Falco confirms that the fix has been applied and shows a correctly captioned image.
Following this, Ethsim2 asks how to refresh existing captions to reflect recent changes and notes that the AI incorrectly identified the Discourse UI as a chatbot interface. sk-or-v1-contents suggests rebuilding HTML, but NateDhaliwal clarifies that this only rebakes oneboxes, not AI captions, suggesting re-uploading might be necessary. sam adds that he is currently reworking the subsystem and will provide toggles to rebake captions in the future. Finally, nat closes the topic, noting that the original issue has been resolved and commenting on the age of the post.
Is the AI Caption guy feeling okay?
A vibrant, bustling street scene in Morocco showcases a spice market with colourful pyramids of spices, merchants attending to customers, and traditional architecture creating a lively atmosphere. (Captioned by AI)
A vibrant red double-decker bus drives along a bustling city street, with pedestrians and buildings visible in the background. (Captioned by AI)
From this post - Email reply formatting - #7 by JammyDodger
Oops, doing a round of LLM upgrades and forgot to set this model as "vision_enabled": true 
Thanks for the report!
how do i refresh this spectacle of a caption given the recent change
also noticed it calling the Discourse UI a chatbot interface
Maybe ask an admin to rebuild the HTML? (don’t know)
IIRC that only rebakes the oneboxes and other data (eg category links), not the AI captions. Perhaps re-uploading would be better.
Also, this topic is more than a year old?
@nat is reworking this subsystem at the moment … and yes we will provide toggles to rebake captions
I’m closing this topic as OP has been rectified.
New issues can be created in new topics.
