I think this is resolved, in that we pulled the feature for now.
Automatic captioning needs a rethink and we needed a breather to rethink the feature.
It is a wonderful feature for search, it is great to be able to type with:images cat and find all the cat pictures.
However the collateral damage here of causing confusing and over long captions is not worth it. We are thinking of ways to enhance the metadata (and handle history) without impacting the user experience.
Additionally, this “extra caption” can be very useful for visually impaired users, so when we revisit we need to take special care with screen readers.